AI fragmentation-based optimization of Sorafenib derivatives targeting VEGFR2 for angiogenesis-related pathologies: a structure-based in-silico study.

Skip to main content An official website of the United States government Here's how you know Here's how you know Official websites use .gov A .gov website belongs to an official government organization in the United States. Secure .gov websites use HTTPS A lock ( Lock Locked padlock icon ) or https:// means you've safely connected to the .gov website. Share sensitive information only on official, secure websites. Search Log in Dashboard Publications Account settings Log out Search… Search NCBI Primary site navigation Search Logged in as: Dashboard Publications Account settings Log in Search PMC Full-Text Archive Search in PMC Journal List User Guide PERMALINK Copy As a library, NLM provides access to scientific literature. Inclusion in an NLM database does not imply endorsement of, or agreement with, the contents by NLM or the National Institutes of Health. Learn more: PMC Disclaimer | PMC Copyright Notice Sci Rep . 2026 Mar 3;16:11848. doi: 10.1038/s41598-026-41232-9 Search in PMC Search in PubMed View in NLM Catalog Add to search AI fragmentation-based optimization of Sorafenib derivatives targeting VEGFR2 for angiogenesis-related pathologies: a structure-based in-silico study Deniz Inan Deniz Inan 1 Özel Nev Hospital, Sanlıurfa, Türkiye Find articles by Deniz Inan 1, ✉ , Sinan Karageçili Sinan Karageçili 1 Özel Nev Hospital, Sanlıurfa, Türkiye Find articles by Sinan Karageçili 1 , Nouman Ali Nouman Ali 2 Department of Molecular Biosciences, Faculty of Physical and Biological Sciences, Rashid Latif Khan University, Lahore, Punjab Pakistan 3 Département of Biotechnology, Faculty of Science and Technology, University of Central Punjab, Lahore, Punjab Pakistan Find articles by Nouman Ali 2, 3, ✉ Author information Article notes Copyright and License information 1 Özel Nev Hospital, Sanlıurfa, Türkiye 2 Department of Molecular Biosciences, Faculty of Physical and Biological Sciences, Rashid Latif Khan University, Lahore, Punjab Pakistan 3 Département of Biotechnology, Faculty of Science and Technology, University of Central Punjab, Lahore, Punjab Pakistan ✉ Corresponding author. Received 2025 Nov 24; Accepted 2026 Feb 18; Collection date 2026. © The Author(s) 2026 Open Access This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License, which permits any non-commercial use, sharing, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if you modified the licensed material. You do not have permission under this licence to share adapted material derived from this article or parts of it. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by-nc-nd/4.0/ . PMC Copyright notice PMCID: PMC13065851  PMID: 41776206 Abstract Vascular endothelial growth factor receptor 2 (VEGFR2) is a central regulator of angiogenesis and endothelial signaling and represents a validated therapeutic target in multiple angiogenesis-associated pathologies. Sorafenib is a clinically used multi-kinase inhibitor with VEGFR2 activity; however, its scaffold is limited by suboptimal pharmacokinetics and toxicity liabilities. In present investigation, we have engineered and evaluated Sorafenib derivatives following artificial intelligence through structure-based drug design (SBDD) and fragments-based drug design (FBDD). Out of twenty derivatives generated, the most efficient derivative, named AI- Fragmented Derivative 7 (Grow), displayed higher binding affinity for the VEGFR2 when compared to Sorafenib. Molecular dynamics simulation was performed to confirm structural stability for the VEGFR2-AI lead complex. Density Functional Theory (DFT) computations showed this by producing a lower HOMO-LUMO energy gap for lead than Sorafenib. Pharmacokinetic profiling (ADMET) associated the compound with enhanced aqueous solubility and absence of hepatotoxicity was predicted compared to the hepatotoxic profile of Sorafenib. Free energy estimations with MM/GBSA and MM/PBSA further supported the thermodynamic properties of derivative 7. Collectively, these computational results support Derivative 7 as a promising VEGFR2-directed lead candidate for angiogenesis-associated diseases; however, all conclusions are based on in silico modeling and require experimental validation of potency, selectivity, and safety. Keywords: VEGFR2, Angiogenesis, Sorafenib, Molecular dynamics, In silico , DFT Subject terms: Computational biology and bioinformatics, Drug discovery Introduction Pathological dysregulation of angiogenesis contributes to a wide spectrum of human diseases. Vascular endothelial growth factor (VEGF) signaling through its primary receptor, VEGFR2, plays a central role in endothelial cell proliferation, migration, survival, and vascular permeability 1 . Aberrant VEGFR2 activation has been implicated in tumor progression, ocular neovascular disorders such as age-related macular degeneration, inflammatory angiogenesis, and fibrotic remodeling 2 . Consequently, VEGFR2 has emerged as a validated and druggable therapeutic target, and multiple small-molecule inhibitors with anti-angiogenic activity have been developed and clinically approved 3 . In addition to its role in oncology and other angiogenesis-associated pathologies, VEGF–VEGFR2 signaling is also involved in physiological vascular development 4 . Disruption of angiogenic balance has been linked to pathological vascular remodeling and endothelial dysfunction (Shibuya, 2010). However, the molecular mechanisms underlying these conditions are complex and involve alterations in VEGF bioavailability and downstream signaling regulation 1 . Therefore, modulation of VEGFR2 activity must be considered within a broader mechanistic and safety framework, particularly in sensitive physiological contexts. Despite the availability of several VEGFR2-targeted agents, many first-generation inhibitors, including multi-kinase compounds such as Sorafenib, are associated with limitations such as off-target kinase activity and dose-limiting toxicities 3 . These challenges underscore the continued need for rational scaffold optimization to improve binding complementarity and pharmacological properties while preserving effective VEGFR2 engagement 2 . One promising molecular target in this regard is vascular endothelial growth factor receptor 2 (VEGFR2). VEGFR2 is a receptor tyrosine kinase that plays a central role in the process of angiogenesis through mediating the effects of VEGF-a in endothelial cells. In fact, virtually all known pro-angiogenic signaling of VEGF is through VEGFR2 while the alternate receptor (VEGFR1) of VEGF plays mostly a decoy or modulatory role 5 . During normal placental development the anti-angiogenic signaling molecules known as vascular endothelial growth factor (VEGF) and related placental growth factor (PlGF) interact with the VEGFR2 on the endothelium cells of the placenta promoting the development of new blood vessels ensuring adequate maternal-fetal circulation 5 . This signaling axis between the proangiogenic molecule, vascular endothelial growth factor (VEGF), and its receptor, vascular endothelial growth factor receptor-2 (VEGFR2), is of the greatest importance in the maintenance of placental vascular homeostasis. Correspondingly, the presence of disorders of the angiogenesis associated with blood vessels (VEGF/VEGFR2 pathways) are implicated in disorders of placental angiogenesis. In pre-eclampsia, an anti-angiogenic factor called soluble primary placental vasculature (VEGF receptor-1) over-produces, bound to the placenta and preventing its activating factors (VEGF and PlGF) from activating VEGFR2. The resulting loss of the VEGFR2 signaling leads to endothelial dysfunction, maternal hypertension, and placental hypoperfusion 5 . Similar angiogenic imbalances are seen in other conditions of placental insufficiency. VEGFR2 is a central receptor tyrosine kinase governing VEGF-driven angiogenesis and endothelial signaling 1 . Dysregulated VEGFR2 signaling contributes to pathological angiogenesis in several clinically important conditions, including solid tumors, ocular neovascular disorders, chronic inflammatory diseases, and fibrotic remodeling 2 . Consequently, VEGFR2 has remained a validated and highly druggable therapeutic target, and multiple small-molecule inhibitors have been clinically approved 3 , 6 . However, many first-generation VEGFR2 inhibitors, including Sorafenib, are limited by multi-kinase promiscuity and dose-limiting toxicities 3 , 6 . Therefore, rational optimization of VEGFR2-targeted scaffolds remains a relevant goal for improving drug-like behavior while preserving potent VEGFR2 engagement 2 . Although VEGF–VEGFR2 signaling is also involved in vascular development and tissue angiogenesis, the present study does not propose VEGFR2 inhibition as a validated therapy for pregnancy-associated disorders 1 . Any potential relevance to placental angiogenesis would require extensive experimental validation in appropriate maternal–fetal safety models and is mentioned only as a long-term translational consideration. In a bid to find treatments for VEGFR2, researchers have looked to oncology, where existing VEGFR2 inhibitors are used as starting points. Sorafenib is a well-known small molecule inhibitor of VEGFR2 that has proven the concept of targeting the tyrosine kinase activity of this receptor. Sorafenib is a multi-kinase inhibitor (a bi-aryl urea compound), and was first approved as a targeted anti-cancer drug in 2007 as a first-line treatment for advanced patients with hepatocellular carcinoma, and subsequently it was approved for advanced renal cell carcinoma and advanced thyroid carcinoma 7 . Mechanistically, sorafenib occupies binding sites for ATP in VEGFR2 (and multiple other kinases such as VEGFR1/3, PDGFR, c-Kit and Raf) and prevents the downstream signaling of the receptor, that is, cell proliferation and angiogenesis in endothelial cells 7 . In tumors this anti-angiogenic effect results in reduced tumor vascularization and tumor shrinkage which confirmed VEGFR2 as a druggable target. However, sorafenib itself is far from being the ideal drug for placental disorders, with substantial limitations in terms of safety and specificity. As a very promiscuous kinase inhibitor, sorafenib has the potential to inhibit more than a dozen off-target kinases, which is a major reason for severe toxicities in patients. Of note, its potent anti-VEGF effect induces cardiovascular side effects - hypertension, vascular dysfunction, cardiac ischemia - that are reminiscent of an exaggerated pre-eclampsia-like state (Blockage of VEGF in healthy tissues induces hypertension) 7 . Sorafenib treatment is also linked to dermatologic toxicity (hand-foot syndrome), gastrointestinal distress, and a narrow therapeutic index. Additionally, cancers often acquire resistance to sorafenib via many different mechanisms, which reduces its long-term effectiveness 8 . Even next generation multitarget VEGFR2 inhibitors (regorafenib, etc.) have not fully resolved these problems as they suffer from moderate clinical benefits and adverse effects related to the mechanism 7 . These limitations render sorafenib unsuitable for direct use in pregnancy but the molecular scaffold and the VEGFR2 binding profile of sorafenib present a useful starting point for rational drug design of improved VEGFR2 inhibitors. By modifying the structure of Sorafenib, there is potential to improve VEGFR2 binding complementarity and optimize predicted drug-like properties. However, systematic kinase selectivity and off-target liability cannot be concluded from VEGFR2-only modeling and must be evaluated through future multi-kinase computational profiling and experimental validation. To design such optimized compounds, we take advantage of improvements in structure-based drug design (SBDD) and associated computational methods. SBDD has become a mainstay of modern drug discovery - providing a rational framework to design and refine drug candidates based on the three-dimensional structure of the target protein 9 . Knowledge of the high-resolution crystal structure of VEGFR2 (including co-crystal structures with inhibitors) allows us to model interactions of new chemical modifications with the VEGFR2 kinase domain at the atomic level. For example, this approach enables an intelligent optimization of binding affinity and selectivity by means of specific hydrogen bonds or hydrophobic pockets on the active site of VEGFR2 instead of relying purely on trial-and-error chemistry. Computer-aided drug design (CADD) techniques, such as molecular docking and virtual screening, further accelerate this process by evaluating large libraries of compounds in silico in order to predict which will experience favorable binding with VEGFR2. These computational tools have dramatically improved the efficiency of lead discovery and optimization, and have contributed to dozens of FDA approved drugs over the last few years. Several small-molecule VEGFR2 inhibitors, including multi-kinase agents such as Sorafenib, Sunitinib, and Pazopanib, have been approved for the treatment of various malignancies and angiogenesis-driven diseases 10 . While these agents demonstrate clinical efficacy through potent VEGFR2 inhibition, their therapeutic use is often limited by off-target kinase activity, dose-limiting toxicities, and restricted cancer-type specificity. Recent computational and fragment-based studies have highlighted the potential of rational molecular optimization to improve VEGFR2 binding characteristics while minimizing adverse effects associated with broad-spectrum kinase inhibition 11 . In this context, fragment-based and AI-driven CADD strategies provide a cost-effective platform to explore structure–activity relationships and guide the design of optimized VEGFR2-targeted candidates prior to experimental validation. Complementing SBDD and CADD, we also use fragment-based drug design (FBDD) in our strategy. FBDD consists of a series of screens of extremely small size molecular fragments (typically, < 300 Da), that weakly bind to the target, followed either by the chemical synthetic growth of the fragment to form fuller size ligands, or the linking of the fragments together to create a fuller structure. This approach is powerful for exploring new chemical space and has produced some successful drugs. In fact, various therapeutic agents have been approved by the FDA that have derived from fragment-based strategies, for example the B-Raf inhibitor vemurafenib (approved to treat melanoma) was developed by optimizing a fragment of low molecular weight to a potent kinase inhibitor 12 . FBDD paradigms are particularly relevant for kinase targets such as VEGFR2, where fragment-based scaffold modification can refine interaction geometry within established binding pockets and optimize pharmacophoric complementarity. By combining motifs derived from the fragments with the sorafenib core scaffold, we hope to find hybrid compounds that exhibit enhanced inhibition of VEGFR2 and exhibit stronger drug-like properties. Overall, the combination of SBDD, CADD and FBDD is a cutting-edge method of drug discovery especially suited for the design of next clinical generation VEGFR2 inhibitors focused on the particular challenges of placental angiogenesis disorders. Placental angiogenesis disorders comprise a heterogeneous group of conditions that extend beyond simple angiogenic insufficiency and include pathological or dysregulated angiogenic states. In several placental pathologies, excessive or aberrant activation of the VEGF–VEGFR2 axis has been associated with endothelial dysfunction, oxidative stress, inflammatory signaling, and abnormal vascular architecture. Consequently, therapeutic strategies in such settings focus on restoring angiogenic balance rather than indiscriminately enhancing angiogenesis. Our study is focused on the rational design of novel VEGFR2-targeted drug candidates relevant to angiogenesis-associated pathologies. Specifically, we applied an AI-guided fragmentation workflow to generate Sorafenib-derived analogues and evaluated their predicted VEGFR2 binding behavior using docking, molecular dynamics simulations, binding free energy estimation, ADMET profiling, and DFT descriptors. The goal of this work was to computationally prioritize a VEGFR2-directed lead candidate for future experimental validation rather than to claim improved kinase selectivity or clinical applicability. Using a combination of structure-based modelling, in silico screening and fragment growing methods we designed a set of candidate compounds and tested their binding interactions against the VEGFR2 kinase domain. This research aims to contribute to the development of an effective drug based on VEGFR2, which would overcome the limitations of current treatments and improve maternal-fetal results of pre-eclampsia and similar conditions. Methodology Retrieval of vascular endothelial growth factor receptor 2 (VEGFR2) 3D structure and validation The three-dimensional coordinates of VEGFR2 were obtained from RCSB Protein Data Bank ( https://www.rcsb.org ) using PDB code 3WZE 13 , 14 . The dataset was downloaded in PDB for compatibility with the following computational workflows. Acquisition of an experimentally resolved structure of VEGFR2 is critical, as the geometry of the receptor is essential for reliable docking, molecular simulations and interaction analyses. PDB ID 3WZE represents VEGFR2 in an inhibitor-bound inactive (type-II) kinase conformation compatible with Sorafenib binding. Since Sorafenib is a type-II VEGFR2 inhibitor that occupies the DFG-out pocket, this structure was selected to ensure mechanistic consistency for docking and scaffold-based derivative evaluation. Following retrieval, the structural integrity of the protein model was assessed with the SAVES v6.1 server ( https://saves.mbi.ucla.edu ) which aggregates several validation tools. The ProCheck module was used to measure the distribution of the back-bone dihedral angles by Ramachandran plot; thus, in order to verify the stereochemical reliability of the model 15 . ERRAT analysis further analyzed non-bonded atomic interaction patterns to give another level of structural verification 16 . Structural validation is vital because it ensures that there are no serious inconsistencies of stereochemistry, which may adversely affect the accuracy of the docking or lead to the generation of false predictions. Protein-protein interaction (PPI) networking of VEGFR2 The interaction landscape of VEGFR2 was queried using the STRING database version 12.0 ( https://string-db.org ) 17 , 18 . The amino acid sequence of VEGFR2 was sent to the STRING interface to produce a high confidence interaction network based on experimentally validated associations, curated pathway information, as well as computational predictions. During analysis the confidence score was set to the default parameters of the STRING to maintain the reproducibility of the analyses and to ensure biologically relevant interactions. Construction of a PPI network is relevant because VEGFR2 functions in the context of a much more general regulatory process; the identification of VEGFR2’s interacting partners puts its role in the angiogenic pathways in perspective and also helps identify proteins that could affect ligand binding or receptor activation, which in turn affects dockings and molecular dynamics outcomes. Retrieval of Sorafenib 3D conformer and AI fragmentation of Sorafenib The 3-dimensional Conformer Sorafenib was downloaded from the PubChem database (PubChem ID: 216239, https://pubchem.ncbi.nlm.nih.gov ) with SDF format 19 , 20 . The structure was converted to PDB in order to be compatible with docking programs and the SMILES notation was retrieved to be used in ADMET and cheminformatics analysis. Representations of the compound in various formats provided seamless integration between computational tools as well as reproducibility of methodology. AI-driven ligand fragmentation and derivative generation was done using the GENERATIVE module at PlayMolecule platform ( https://open.playmolecule.org/tools/generative ) 21 , 22 . This tool applies the Chemically Reasonable Mutations (CReM) framework, which utilizes rule-based functional group replacement and fragment growth from a library of synthetically accessible fragments. Two modes of mutation, Replace and Grow, were used, and explored structural diversity without compromising synthetic feasibility and chemical integrity. Generated derivatives were visually scrutinized for chemical stability before implementation into the docking screen. This AI-based methodology is essential and paves the way for rapid, systematic growth of the chemical space around a known inhibitor for more chances of discovering improved VEGFR2 binders that may be more resistant to manual modification or traditional medicinal chemistry attempts. The mutation radius was restricted to conservative growth parameters within the CReM framework to preserve scaffold integrity, maintain synthetic feasibility, and avoid excessive molecular enlargement that could compromise ligand efficiency or steric compatibility within the VEGFR2 ATP-binding pocket. Although the fragmented derivatives were computationally generated from CReM, algorithms of these transformations are based on synthetically accessible fragments and validated substitution rules. Consequently, the analogs designed are not theoretical structures but rather structures that can be synthetically prepared in the laboratory via the traditional routes of organic chemistry. CReM imposes valence rules, aromaticity constraints, and chemically reasonable mutations in order to ensure that generated compounds lie in the domain of practical medicinal chemistry. Prediction of VEGFR2 structure active sites Although VEGFR2 has been crystallized in complex with Sorafenib (PDB ID: 3WZE), active-site prediction was performed using the machine learning–based PrankWeb server ( https://prankweb.cz ) to independently identify and validate druggable binding cavities in a ligand-independent manner 23 . This approach was adopted to avoid bias arising solely from the presence of a co-crystallized ligand and to ensure consistency with the AI-driven computational framework of the present study. The processed structure of VEGFR2 was uploaded to PrankWeb to obtain ranked pockets along with scores, residues composition and three-dimensional coordinates. The high scoring pocket and was called pocket 1 which was predicted for molecular docking on the basis of high probability score, as well as related to the functional domain of the receptor. Accurate determination of ligand binding areas is crucial since it allows the simulation of docking to biologically meaningful cavities rather than nonfunctional surface area areas to make predicted interactions more reliable and builds confidence that in silico events are likely to correspond to plausible receptor-ligand recognition in vivo. Preparation of receptor protein The structure of VEGFR2 was pre-treated for possible docking, removing crystallographic water molecules, heteroatoms and co-crystallized ligands with BIOVIA Discovery Studio to avoid non-essential components causing any perturbations in ligand binding 24 . The cleaned structure was subjected to structural correction using PDBFixer which is a module in the SAMSON Suite to rectify missing atoms, close chain breaks and correct steric clashes, in order to improve structural completeness 25 . Subsequent conservative energy minimization from SwissPDB Viewer refined the bond geometries and reduced residual strain as well as maintaining the native fold of the kinase domain 26 . The finalized minimized structure was saved as a PDB for direct use in docking. Proper receptor preparation is imperative, as inaccuracies, e.g. the missing residues or the unsolved steric clashes, can significantly influence the orientation of the ligand and affinity estimations, and thus compromise the reproducibility and reliability of downstream docking and simulation studies. Molecular docking and interaction studies of AI fragmented derivatives with VEGFR2 Docking was done by AutoDock vina extended within the SAMSON Suite in order to assess the binding affinities and interaction profile of all AI fragmented Sorafenib derivatives against VEGFR2 27 . Before docking, each ligand was energy minimized using the Universal Force Field to remove unfavorable contacts and produce a stable conformation. The structure of the VEGFR2 was protonated in SAMSON Suite, and the docking grid box was centered to Pocket 1 (predicted by PrankWeb) at coordinates X = 24.2 Å, Y = 21.5 Å and Z = 34.8 Å and dimensions X = 20 Å, Y = 20 Å, Z = 20 Å. Exhaustiveness was set to 8, trading off computational efficiency and sampling depth 28 . Results were separately reproduced in PyRx under identical parameters in order to verify reproducibility across platforms 29 , 30 . Protein-ligand interactions were investigated further using the SAMSON interaction module, and the PLIP server ( https://plip-tool.biotec.tu-dresden.de/ ) to create detailed two and three-dimensional interaction diagrams, hydrophobic contacts, together with hydrogen-bonding 31 , 32 . Docking allows a first-order estimation of the binding propensity of each derivative as well as candidates that can form stable, biologically relevant interactions in the experimentally relevant active site, thus forming the basis of subsequent molecular dynamics and quantum mechanical studies. AutoDock Vina was selected due to its balance of computational efficiency and reliable pose prediction for kinase–ligand systems. Screening of AI derivatives based on ADMET analysis Pharmacokinetic and physicochemical properties of the compounds were calculated using the SwissADME server ( https://www.swissadme.ch ) 33 . The SMILES representation was submitted to generate predictions of absorption, distribution, metabolism and excretion (e.g. Lipophilicity, aqueous solubility, gastrointestinal absorption, blood-brain barrier permeability, cytochrome P450 interactions, drug-likeness filters) (Lipinski, Veber, Ghose, Egan). This step is a very important one because ADME profiling in the early stages of drug discovery gives an indication whether a compound has favorable attributes that are necessary to proceed as a drug candidate. Toxicity predictions were carried out using Deep-PK ( https://biosig.lab.uq.edu.au/deeppk/ ), in which the chemical structure was analyzed with deep learning models differed based on toxicological data sets 34 . Endpoints such as mutagenicity, carcinogenicity, organ specific toxicity and environmental toxicity indicators were tested. Incorporation of in silico ADMET and toxicity profiling is indispensable and an important source of early estimates on safety and pharmacokinetic liabilities to inform the rational choice of lead compounds before resource-intensive experimental validation. ADMET predictions were conducted using widely used web-based platforms to obtain rapid, consensus-level pharmacokinetic screening. Pharmacophore characterization and density functional theory (DFT) of lead drug candidate Pharmacophoric features of the selected lead compound were explained by using the Pharmit server ( http://pharmit.csb.pitt.edu ) 35 . The three-dimensional structure of the ligand was uploaded to automatically identify the presence of hydrogen bond donors and acceptors, aromatic centers and regions of hydrophobic interactions to help clarify the roles they play in potential receptor binding. This analysis highlights the chemical determinants necessary for VEGFR2 interaction and describes how mode of modification of structure affect biological activity. Quantum-chemical analysis of the lead compound was also carried out using the Gaussian 09 W at the B3LYP/6-31G(d, p) level of theory to optimize the molecular geometry and get the lowest energy conformation 36 . Frontier orbital energies - including HOMO and LUMO were computed in order to determine electronic distribution, and HOMO - LUMO gaps reflecting chemical hardness and charge transfer propensity. Molecular electrostatic potential (MEP) mapping was used to visualize electron rich and electron deficient sites to identify areas prone to electrophilic/nucleophilic attack. DFT analysis provides a detailed insight into electronic behavior, structural stability and reactivity which, when complemented with pharmacophore evaluation, contribute to rational drug design. Molecular dynamics simulation of lead drug candidate with VEGFR2 Molecular dynamics simulations were conducted for the top-ranked derivative selected through a multi-parameter screening strategy to assess binding stability and interaction persistence under dynamic conditions rather than to perform comparative ranking among multiple docked ligands. Dynamic behavior stability or persistence interaction of the lead AI - fragmented derivative in the VEGFR2 binding pocket was studied by molecular dynamics simulations under near-physiological conditions. The VEGFR2–ligand complex selected from docking analysis was simulated using OpenMM using AMBER forces: protein parameters were defined with ff19SB while ligand parameters were generated via the workflow compatible with Antechamber 25 , 37 . The system was solvated in explicit SPC water within an orthorhombic box, keeping a minimum 10 microns buffer of the complex to periodic boundaries. Sodium and chloride ions were added to neutralize the system and to simulate the physiological ionic strength (0.15 M) 38 . Prior to production the energy of the system was minimized in order to eliminate steric clashes, and the system was gradually heated and equilibrated under the NVT and NPT ensemble at 300 K and 1 atm, respectively 25 . SHAKE constraints were imposed on hydrogen bearing bonds which allowed a 2-fs integration timestep. Long-range electrostatics were calculated using Particle Mesh Ewald with a 10Å real-space cutoff; Lennard-Jones interactions were truncated at 10Å with a switching function starting at 9Å to achieve smooth decay of the forces. The production run lasted for 500 ns under periodic boundary conditions, with trajectory snapshots taken every 10 ps for later structural, energetic and conformational analysis. Population-based stability and conformational persistence were evaluated through time-dependent distributions of RMSD, RMSF, SASA, and hydrogen bond counts rather than separate probability density function plots. Binding free energies have been estimated using both MM/GBSA and MM/PBSA methods, estimated from 5,000 uniformly sampled snapshots from the trajectory in the last segment of the 500 ns trajectory. In the present study, PCA was employed to characterize dominant collective motions of the protein–ligand complex. Construction of a PCA-based free energy landscape was not performed, as the analysis was intended to qualitatively assess conformational stability rather than derive energetic basins. Incorporating molecular dynamics provides a time-resolved view of receptor-ligand interactions and can be used to examine stability, conformational flexibility, and energetic favorability in addition to static docking predictions. Molecular dynamics simulations were performed using the OpenMM simulation engine, which was selected for its high computational efficiency, flexibility, and reproducibility in biomolecular simulations, as well as its extensive validation in protein–ligand interaction studies. Justification for computational tool selection Multiple computational tools are available for molecular docking, molecular dynamics simulations, binding free energy calculations, and ADMET prediction. In the present study, tool selection was guided by methodological compatibility, widespread validation in the literature, and suitability for the specific computational task rather than comparative benchmarking. The employed tools represent well-established and extensively cited platforms for structure-based drug design and have been successfully applied in studies involving kinase targets and small-molecule inhibitors. Given the scope of this work, alternative software packages with overlapping functionality were not explored in parallel, as the objective was to generate a coherent and internally consistent in silico pipeline rather than perform inter-software performance comparisons. Results Retrieval of VEGFR2 3D structure and validation The three-dimensional structure of VEGFR2 was retrieved using the crystallographic model with PDB ID 3WZE, which represents the kinase domain of the receptor in complex with an inhibitor (Fig. 1 A). The structure is a member of the classification: TRANSFERASE/TRANSFERASE INHIBITOR and it is from Homo sapiens and expressed in Spodoptera frugiperda . The model was solved by X-ray diffraction data with a resolution of 1.90 Å, which yielded high-quality atomic coordinates for computational analysis. The protein is also said to contain engineered mutations introduced during the experimental preparation, the entry reports. Fig. 1. Open in a new tab ( A ) VEGFR2 3D structure, ( B ) Ramachandran plot, and ( C ) ERRAT quality factor analysis. Quality assessment on the structure revealed good stereochemical reliability. Using the Ramachandran plot, it was found that 93.4% of residues were located in the most favored regions, 6.2% were located in additional allowed regions, 0.4% were located in generously allowed regions, and no residues were located in disallowed regions (Fig. 1 B). All 258 non-glycine and non-proline residues were accounted for, indicating good backbone geometry with no problematic torsional angles. The ERRAT analysis further backed up the structural integrity with an overall quality factor of 98.540 which indicates a high-reliability nonbonded atomic interaction profile (Fig. 1 C). Together, these validation parameters show the retrieved structure of VEGFR2 has the quality of accuracy and stereochemical consistency that is required to support docking, molecular dynamics, and interaction analysis. Protein-protein interaction (PPI) networking of VEGFR2 Analysis of the VEGFR2 interaction landscape revealed a dense interconnected protein-protein interaction (PPI) network, with 11 nodes and 45 edges, indicating wide-ranging functional associations among the constituent proteins (Fig. 2 ). The average node degree was 8.18, and the average local clustering coefficient was 0.884, which indicates the proteins form a highly interconnected regulatory module, in contrast to isolated pairwise interactions. The number of edges observed was significantly higher than the 18 expected in random models of networks, with a p-value for enrichment of 1.03 * 10 − 7, thus confirming that the proteins in this network are biologically interconnected and participate in common signaling pathways rather than stochastic interactions in the network. Fig. 2. Open in a new tab Protein–protein interaction (PPI) network of VEGFR2 retrieved from String V12.0. Functional annotation of the interaction partners revealed enrichment of angiogenesis-associated and inflammation-related signaling components, consistent with the established biological role of VEGFR2 in vascular development and endothelial regulation. Several proteins within the network are linked to transcriptional regulation and cellular stress responses, reflecting the integration of VEGFR2 signaling within broader regulatory cascades. Importantly, this PPI analysis was performed solely to provide pathway-level biological context for VEGFR2 and to confirm its central connectivity within angiogenic signaling networks. No proteins identified in the PPI network, including transcriptional regulators, were subjected to molecular docking, molecular dynamics simulation, or structure-based computational analysis in this study. All subsequent computational modeling, including docking, molecular dynamics simulations, binding free energy calculations, ADMET profiling, and DFT analysis, was performed exclusively for VEGFR2. AI fragmentation of Sorafenib derivatives The structural diversification of Sorafenib resulted in a focused library of twenty chemically valid derivatives that were developed by a combination of fragment replacement and fragment growth strategies (Table 1 ). Ten analogs were made using the Grow approach, adding other substituents to modifiable sites of the Sorafenib scaffold, and the remaining ten analogs were synthesized through the targeted replacement of specific functional groups with chemically compatible fragments. Each derivative maintained the basic pharmacophoric structure of Sorafenib, providing the starting point for preserving essential inhibitory motifs and enabling the investigation of an expanded chemical space. Table 1. AI-generated Sorafenib derivatives from Grow and Replace strategies with their structural representations. Open in a new tab Visual inspection of generated structures proved the synthetic feasibility of these entities, as all the analogs showed the preservation of valence, correct bonding patterns and chemical environments. None of the resulting derivatives showed steric clashes, unstable motifs or reactive functionalities which could affect the molecular integrity. This observation supports the idea that chemical mutations guided by AI create molecules that lie within the synthetically attainable chemistry boundaries. The full series of derivatives represents a balanced and varied chemical series with variations in the aromatic substituents, hydrogen-bonding features, hydrophobic extensions and heteroatom incorporation. This structural diversity lays a solid basis for comparative docking and subsequent computational evaluation in order to provide more potent VEGFR2 inhibitors compared to the parent compound. Prediction of VEGFR2 active sites The study of the structure of VEGFR2 identified three candidate ligand binding sites, out of which Pocket 1 showed itself to be the most prominent and biologically relevant cavity. Pocket 1 had the highest score 29.43 with a probability of 0.914, meaning that there is a strong possibility of it being the true functional binding region of the receptor. Pocket 1 corresponded to the canonical ATP-binding site of VEGFR2, which is the experimentally validated binding site of Sorafenib in the co-crystal structure (PDB ID: 3WZE). This pocket consisted of 38 amino acideresidues that formed a structurally well-defined and spacious cavity with dimension of around 24.26 Angstrom (X), 21.54 Angstrom (Y), and 34.87 Angstrom (Z) (Fig. 3 ; Table 2 ). Its high average conservation score of 2.064 also supports its importance indicating that residues in this region are evolutionarily conserved and may be important for ligand recognition or catalytic control. The general features of Pocket 1 make them a suitable place to accommodate the ligand, thereby leading to their choice for all the following docking and interaction studies. Fig. 3. Open in a new tab Predicted active sites of VEGFR2 with Pocket 1 highlighted as the highest-ranked binding cavity. Table 2. Predicted binding pockets of VEGFR2 with scores, probabilities, residue counts, conservation values, and pocket dimensions. No. Score Probability No. of Residues Avg Conservation Dimensions X Y Z 1 29.43 0.914 38 2.064 24.2646 21.538 34.8743 2 2.03 0.043 7 1.185 14.7395 20.9659 37.4357 3 0.98 0.006 7 1.175 9.1365 20.993 29.1983 Open in a new tab There were two additional pockets identified and incorporated as subsets of the comparative evaluation. These pockets had lower scores, lower residue counts and lower probability values, indicating lower chances of being a primary ligand binding centers. Their numerical details are presented in the corresponding results table, so that a complete overview of all predicted cavities can be obtained, thus highlighting Pocket 1 as the most functionally significant location for inhibitor engagement. The concordance between the ML-predicted Pocket 1 and the known ATP-binding site further supported its selection for subsequent molecular docking and molecular dynamics simulations. Molecular docking and interaction studies of AI-fragmented derivatives with VEGFR2 The docking evaluation of the original drug and of all twenty of the compounds that were fragmented with the aid of AI showed distinct variations in binding performance at the active site of the target protein (VEGFR2). The reference compound, sorafenib, was chosen as it gave a value for binding affinity of -10.2 kcal/mol, a value that thus serves as the benchmark against which the improvement gained by modification of the molecule can be compared. And of the brand-new entities, the one with the best affinity, AI - Fragmented Derivative 7 (Grow) had a binding energy of -11.7 kcal/mol across all the docking platforms, so far superior to Sorafenib. On the other hand, the weakest interacting compound, AI- Fragmented Derivative 5 (Grow), showed an affinity of -8.3 kcal/mol, indicative of a reduced complementarity with the binding pocket in VEGFR2. The binding affinities for all ligands in complex with VEGFR2 are shown in Table 3 . Because the generated derivatives differ only by minor structural modifications relative to Sorafenib, ligand size variation was limited; nevertheless, ligand efficiency-based evaluation and multi-ligand MM/GBSA rescoring represent valuable future extensions of this work. Table 3. Binding affinities of Sorafenib and all AI-fragmented derivatives using two independent docking platforms. Sr. Compound PyRx Autodock vina expended RSMD Binding affinity (kcal/mol) RSMD Binding affinity (kcal/mol) 1. Sorafenib 0.00 -10.2 0.00 -10.124 Grow 2 AI Fragmented Ligand 1 0.00 -10.2 0.00 -9.805 3 AI Fragmented Ligand 2 0.00 -11.3 0.00 -10.577 4 AI Fragmented Ligand 3 0.00 -10.3 0.00 -9.829 5 AI Fragmented Ligand 4 0.00 -8.3 0.00 -10.405 6 AI Fragmented Ligand 5 0.00 -8.7 0.00 -8.654 7 AI Fragmented Ligand 6 0.00 -10.4 0.00 -9.466 8 AI Fragmented Ligand 7 0.00 -11.7 0.00 -10.915 9 AI Fragmented Ligand 8 0.00 -11.1 0.00 -9.647 10 AI Fragmented Ligand 9 0.00 -11.7 0.00 -10.726 11 AI Fragmented Ligand 10 0.00 -11.7 0.00 -9.621 Replace 12 AI Fragmented Ligand 1 0.00 -9.0 0.00 -10.100 13 AI Fragmented Ligand 2 0.00 -10.3 0.00 -9.951 14 AI Fragmented Ligand 3 0.00 -10.4 0.00 -9.396 15 AI Fragmented Ligand 4 0.00 -10.8 0.00 -10.194 16 AI Fragmented Ligand 5 0.00 -10.2 0.00 -9.832 17 AI Fragmented Ligand 6 0.00 -10.5 0.00 -9.600 18 AI Fragmented Ligand 7 0.00 -10.3 0.00 -10.349 19 AI Fragmented Ligand 8 0.00 -9.4 0.00 -9.152 20 AI Fragmented Ligand 9 0.00 -10.0 0.00 -9.460 21 AI Fragmented Ligand 10 0.00 -9.1 0.00 -9.803 Open in a new tab It should be noted that although larger ligands may exhibit more favorable docking scores due to increased contact surface, docking score alone was not used as the sole criterion for lead selection. Ligand efficiency is a useful normalization metric for comparing compounds of substantially different sizes; however, in the present study, AI-based fragmentation resulted in only minor structural modifications relative to Sorafenib rather than extensive molecular enlargement. More importantly, lead prioritization was performed using a multi-parameter evaluation strategy incorporating binding pose quality, key residue interactions, molecular dynamics stability, binding free energy calculations, pharmacokinetic profiling, and electronic structure analysis. This integrated assessment was employed to avoid bias associated with any single scoring metric. Detailed interaction analysis of the top performing derivative shows that it creates a strong and supportable binding network within Pocket 1. The ligand makes vast hydrophobic contacts with Val848, Ala866, Leu889, Val899, Val916, Leu1035, and Phe1047, which contributes to deep anchoring within the hydrophobic channel (Table 4 ). Furthermore, multiple hydrogen bonds with Glu885, Val914, Cyt919, Gly922 and Asp1046 provide additional stabilization of the binding conformation. The residue Asp1046, which is part of the activation loop, is involved in two distinct hydrogen bond interactions with the ligand and indicates strong and stable interaction pattern (Table 5 ). The dock pose, as well as the two-dimensional and three-dimensional interactions of the lead drug candidate with the VEGFR2 are visualized in Fig. 4 . None of the generated derivatives, including Derivative 7 (Grow), demonstrated binding to a previously uncharacterized sub-pocket of VEGFR2. All ligands occupied the canonical type-II inhibitor binding region corresponding to the ATP-binding site and adjacent DFG-out pocket observed in the Sorafenib-bound crystal structure (PDB ID: 3WZE). Therefore, the present AI-guided fragmentation approach refined interactions within the established binding cavity rather than identifying a novel allosteric or secondary sub-pocket. Table 4. Hydrophobic interaction profile of AI-Fragmented Derivative 7 within the VEGFR2 binding site. Index Residue AA Distance (Å) Ligand Atom Protein Atom 1 848 A VAL 3.71 4834 547 2 866 A ALA 3.41 4827 810 3 885 A GLU 3.72 4856 1104 4 889 A LEU 3.85 4849 1181 5 899 A VAL 3.87 4844 1343 6 916 A VAL 3.61 4834 1589 7 1035 A LEU 3.82 4827 2680 8 1046 A ASP 3.92 4848 2862 9 1047 A PHE 3.45 4840 2879 Open in a new tab Table 5. Hydrogen bond interactions between AI-Fragmented Derivative 7 and VEGFR2 residues. Index Residue AA Distance H-A  (Å) Distance D-A  (Å) Donor Angle Protein donor? Side Chain Donor Atom Acceptor Atom 885 A GLU 1.92 3.92 106.16 No Yes 4845 [O3] 1107 [O-] 2 914 A VAL 3.23 3.66 108.63 No No 4864 [O3] 1551 [O2] 3 919 A CYS 2.39 3.35 154.79 Yes No 1634 [Nam] 4824 [N3] 4 922 A GLY 3.56 3.95 105.13 Yes No 1687 [Nam] 4819 [N3] 5 1046 A ASP 2.65 3.62 157.99 Yes No 2858 [Nam] 4842 [N3] 6 1046 A ASP 2.41 3.34 151.33 No No 4846 [N3] 2861 [O2] Open in a new tab Fig. 4. Open in a new tab ( A ) Binding pose of AI-Fragmented Derivative 7 (Grow) in VEGFR2, ( B ) two-dimensional interaction diagram, and ( C ) three-dimensional interaction visualization. Although several derivatives exhibited favorable docking scores, selection of derivative 7 for molecular dynamics simulations was not based solely on binding affinity. Instead, derivative 7 demonstrated the most consistent binding pose within the VEGFR2 ATP-binding site, preserved key interactions observed for Sorafenib, and showed favorable physicochemical and pharmacokinetic profiles during preliminary screening. Notably, ADMET screening results for all derivatives are provided in Table 6 to ensure reproducibility and transparency of the lead selection process. These combined criteria supported its prioritization for dynamic and free energy analyses aimed at validating binding stability rather than ranking multiple candidates. In contrast, the weakest binding derivative has lost the extensive network of favorable contacts observed in the lead compound. Its poor interactions with R-group hydrophobic residues and the lack of strong hydrogen bond donors and acceptors result in a less stable pose, as well as a weak predicted affinity. These results highlight the importance of understanding AI-guided changes into the Sorafenib scaffold, to significantly impact the molecular fit and density of interactions, culminating in better embedding of a superior candidate for VEGFR2 inhibitor. Table 6. Overall ADMET comparison of all AI derivatives. AI derivative Lipophilicity (consensus log P o/w ) Water solubility GI absorption BBB permeant Lipinski Grow AI Derivative 1 4.27 Poorly soluble Low No Yes AI Derivative 2 4.57 Poorly soluble Low No Yes AI Derivative 3 3.48 Poorly soluble Low No Yes AI Derivative 4 3.97 Poorly soluble Low No Yes AI Derivative 5 3.78 Poorly soluble Low No Yes AI Derivative 6 4.71 Poorly soluble Low No Yes AI Derivative 7 3.72 Poorly soluble Low No Yes AI Derivative 8 3.92 Poorly soluble Low No Yes AI Derivative 9 3.57 Poorly soluble Low No Yes AI Derivative 10 3.75 Poorly soluble Low No Yes Replace AI Derivative 1 3.72 Poorly soluble High No Yes AI Derivative 2 4.46 Poorly soluble Low No Yes AI Derivative 3 4.41 Poorly soluble Low No Yes AI Derivative 4 5.26 Poorly soluble High No Yes AI Derivative 5 4.32 Poorly soluble High No Yes AI Derivative 6 5.93 Poorly soluble Low No Yes AI Derivative 7 3.62 Poorly soluble High No Yes AI Derivative 8 4.71 Poorly soluble High No Yes AI Derivative 9 4.36 Poorly soluble High No Yes AI Derivative 10 4.51 Poorly soluble High No Yes Open in a new tab Screening of ligands based on ADMET analysis ADMET screening was performed for all twenty AI-generated derivatives to support transparent lead prioritization beyond docking score alone. The comparative ADMET outcomes for the full derivative set are summarized in Table 6 . This screening step was used to identify candidates that retained acceptable drug-likeness and pharmacokinetic feasibility prior to selecting the final lead compound for high-resolution validation. A comparative analysis of ADMET showed that AI-Fragmented Derivative 7 (Grow) exhibits a pharmacokinetic profile and safety characteristics broadly consistent with those of Sorafenib, with the added benefits relevant to optimizing the drug. The lead compound had a slightly higher molecular weight (492.84 g/mol compared to 464.82 g/mol for Sorafenib), which is due to the added structural fragments in the lead compound generated via AI-guided modification. Despite this increment, both molecules met Lipinski’s rule of five without violating any parameters, hence confirming the derivative did not violate the acceptable parameters of drug-likeness. The topological polar surface area (TPSA) of the lead was found to be increased to 109.42 Angstrom squared square compared with the Sorafenib polar surface area which was 92.35 Angstrom squared, demonstrating increased polarity and hydrogen-bonding potential. This observation is in accord with the more extensive interaction network revealed by docking and molecular dynamics simulation. Derivative 7 (Grow) was prioritized as the final lead because it combined one of the strongest docking affinities in the series with a favorable ADMET profile, including improved predicted lipophilicity balance and reduced hepatotoxicity risk relative to Sorafenib. Both compounds showed low gastrointestinal absorption and were predicted to be impermeable across the blood–brain barrier, suggesting similar characteristics of systemic distribution. This inference is supported by the Boiled-egg diagram presented in Fig. 5 . Lipophilicity showed some marginal improvement in the lead (logP 3.72) compared to Sorafenib (logP 4.10) reflecting a slightly more balanced hydrophobic–hydrophilic profile. Water solubility also improved slightly in the derivative (logS = approx. -4.85) compared with Sorafenib (logS = -5.11) and might aid formulation and bioavailability. Cytochrome P450 inhibition profiles were similar for both molecules, the predicted inhibitory effects on major isoforms including CYP1A2, CYP2C19, CYP2C9, CYP2D6 and CYP3A4 being nearly identical. This similarity means that the metabolic pathways of the lead compound are very similar to the metabolic pathways of Sorafenib, implying that the metabolism of the compound will be predictable, but also that care should be taken when considering the potential for drug-drug interaction. Skin permeation (logKp) was slightly lower in the derivative (-6.80 cm − 1 ) relative to Sorafenib (-6.25 cm − 1 ) indicating the increased polarity and reduced passive permeability in the case of the derivative, details of the same have been described during comparative ADME analysis (Table 7 ). Fig. 5. Open in a new tab BOILED-Egg predictive model showing gastrointestinal absorption and BBB permeability. Table 7. Comparative ADME properties of AI-Fragmented Derivative 7 and Sorafenib including physicochemical, pharmacokinetic, and drug-likeness parameters. Parameters AI-fragmented lead Sorafenib Physiochemical properties Formula C22H16ClF3N4O4 C21H16ClF3N4O3 Molecular weight 492.84 g/mol 464.82 g/mol Num. heavy atoms 34 32 Num. from. Heavy atoms 18 18 Fraction Csp3 0.09 0.10 Num. rotatable bonds 10 9 Num. H-bond acceptors 8 7 Num. H-bond donors 3 3 Molar Refractivity 117.87 112.48 TPSA (Å 2 ) 109.42 Å 2 92.35 Å 2 Lipophilicity Log Po/w 3.72 4.10 Water Solubility Log S (ESOL) -4.85 -5.11 Class Poorly soluble Poorly soluble Pharmacokinetics GI absorption Low Low BBB permeant No No P-gp substrate No No CYP1A2 inhibitor Yes Yes CYP2C19 inhibitor Yes Yes CYP2C9 inhibitor Yes Yes CYP2D6 inhibitor Yes Yes CYP3A4 inhibitor Yes Yes Log K p (skin permeation) -6.80 cm/s -6.25 cm/s Drug Likeness Lipinski Yes; 0 violation Yes; 0 violation Ghose No; 2 violations: MW > 480, WLOGP > 5.6 No; 1 violation: WLOGP > 5.6 Veber Yes Yes Egan No; 1 violation: WLOGP > 5.88 No; 1 violation: WLOGP > 5.88 Muegge Yes Yes Bioavailability Score 0.55 0.55 Open in a new tab Toxicity predictions showed an overall similar safety profile between the two molecules (Table 8 ) . Both were found to be safe for mutagenicity, carcinogenicity and important endpoints on endocrine receptors. Nevertheless, lead had an advantage in terms of hepatotoxicity potential of the drug: it was predicted safe in DILI (probability of 0.39) whereas Sorafenib had a tendency of toxicity (0.591). Other toxicity endpoints, including respiratory toxicity, the formation of micronucleus, and environmental toxicity markers were similar between the compounds. Collectively, these ADMET findings position AI- Fragmented Derivative 7 as a pharmacokinetically feasible and safer alternative to Sorafenib and thus support the progression of this molecule as a refined VEGFR2 inhibitor candidate. Table 8. Toxicity predictions for the AI-Fragmented Derivative 7 and Sorafenib across multiple toxicity endpoints. Open in a new tab Pharmacophore characterization and DFT of AI-fragmented lead The pharmacophore analysis of AI-Fragmented Derivative 7 (Grow) revealed the critical functional features accounting for its high binding affinity towards VEGFR2 (Fig. 6 A–D). The lead compound contained 2 aromatic centers, 3 hydrogen bond donors, 4 hydrogen bond acceptors and 6 hydrophobic moieties, which represented a balance of physicochemical features favorable for kinase inhibition. Relative to Sorafenib, the lead retained an equivalent number of hydrogen bond donors and hydrophobic features while adding another hydrogen bond acceptor, presumably enhancing its versatility of interactions within Pocket 1. The slight reduction in aromatic number, from three to two, indicates a more sophisticated scaffold, thus reducing unnecessary rigidity but not compromising binding relevant aromaticity. These pharmacophoric attributes collectively form the basis of the superior interaction network observed during docking and MD simulations. The DFT analysis provided additional information on the electronic properties and reactivity of the derivative. Geometry optimization reached a stable minimum, allowing a reliable evaluation of the frontier molecular orbitals. The calculated HOMO-LUMO energy gap was 0.15433 a.u. (4.200 eV) for the lead compound; which is slightly less than the gap of 0.16219 a.u. (4.414 eV) of Sorafenib (Fig. 7 A–B). This shortening of energy gap suggests that the lead has high electronic polarizability as well as a greater ability to participate in charge transfer interactions in the VEGFR2 active site. The molecular electrostatic potential (MEP) map shows clear electron-rich and electron-deficient regions, in line with the compound’s hydrogen bonding pattern and the stability shown during the MD simulation (Fig. 7 C -D). Collectively, the pharmacophore and DFT results support that AI- Fragmented Derivative 7 (Grow) exhibits a well-organized distribution of functional groups and favorable electronic features in order to complement both the docking and dynamic simulation findings supporting it as the most promising candidate for a VEGFR2 inhibitor. Fig. 6. Open in a new tab Pharmacophore model and structural representation of ( A – C ) Sorafenib and ( B – D ) AI-Fragmented Derivative 7. The DFT-derived electronic properties of derivative 7 provide insight into its potential inhibitory activity against VEGFR2. The HOMO distribution was predominantly localized over the heterocyclic core and linker regions, suggesting electron-donating capability toward the receptor binding site, while the LUMO density was concentrated on electron-deficient regions capable of accepting electron density during ligand–receptor interactions. The moderate HOMO–LUMO energy gap indicates a balance between molecular stability and chemical reactivity, which is favorable for sustained binding interactions. The molecular electrostatic potential (MEP) map revealed distinct electron-rich and electron-poor regions that spatially correspond to hydrogen bond donors/acceptors and hydrophobic interaction sites observed in docking and molecular dynamics analyses. Fig. 7. Open in a new tab Electronic structure and electrostatic potential analysis of Sorafenib and AI-Fragmented Derivative 7 (Grow). ( A ) Highest Occupied Molecular Orbital (HOMO) and Lowest Unoccupied Molecular Orbital (LUMO) distributions of Sorafenib. ( B ) HOMO and LUMO distributions of Derivative 7 (Grow). In the orbital representations, red and green lobes correspond to opposite phases of the molecular wavefunction. ( C ) Molecular Electrostatic Potential (MEP) surface of Sorafenib mapped onto the electron density surface. ( D ) MEP surface of Derivative 7. In the MEP maps, red regions indicate areas of higher electron density (negative electrostatic potential), blue regions indicate electron-deficient areas (positive electrostatic potential), and green to yellow regions represent intermediate electrostatic potential values. Molecular dynamics simulation of AI-fragmented lead with VEGFR2 Root mean square deviation (RMSD) : Analysis of the RMSD of the VEGFR2 - lead compound complex showed a stable trajectory over the full simulation time (500 nanoseconds in length) shown in Fig. 8A . The protein RMSD stabilized near 0.08 nm and exhibited a gradual increase until it stabilized within 0.13–0.18 nm after the first 50–60 ns, indicating a well-equilibrated backbone conformation. The ligand RMSD was significantly lower with a consistent and repetitive value of about 0.05–0.10 nm, suggesting that the compound had a particular binding orientation in the active site. The active-site RMSD showed a similar pattern, stabilizing between 0.12 and 0.17 nm. Collectively, low amplitude variations for all RMSD profiles showed structural stability and tight association of the protein, binding site and ligand were maintained over the 500-nanosecond simulation. Root mean square fluctuation (RMSF) : The RMSF plot showed a low residue wise flexibility observed over the VEGFR2 structure (Fig. 8B). Most residues showed fluctuation within 0.05–0.12 nm, suggesting a compact rigid binding environment. Increased fluctuation was seen in the N-terminal region (up to 0.45 nm) and in specific loop segments near residue 130–140 and 200–210, where fluctuation was high (0.30–0.34 nm). These peaks correspond to solvent exposed or inherently flexible regions and do not compromise ligand binding. Importantly, residues that make up Pocket 1 showed low and stable fluctuations, thereby demonstrating fixed interactions between the ligand and the binding site over the course of the simulation. Radius of gyration (rGyr) : The radius of gyration was stable throughout the duration of the simulation ranging between 1.95 and 2.01 nm (Fig. 8C). Initial slight fluctuations were observed during the first 40–50 ns, which represent possible early equilibration adjustments. Subsequent to this phase, rGyr values stabilized around 1.96–1.98 nm that reflects maintenance of the overall compactness of VEGFR2. The lack of any significant drifts or long-term expansion shows that binding of the lead compound does not cause an unfolding or destabilization of the protein architecture. Solvent accessible surface area (SASA) : Initially, the SASA profile fluctuated at 1420–1500 nm2 and reflects normal breathing motions of the protein surface (Fig. 8D). Following the time of approximately 80–100 ns, the value of SASA showed the existence of mild oscillations within 1440–1520 nm 2 , with no sustained increase or decrease. This type of behavior is consistent with stable solvent and lack of significant collapse or expansion of protein structure upon binding. The small periodic oscillations seen for the 500-nanosecond simulation are inherent solvent-protein dynamics and additional support for the conformational stability of the complex. Fig. 8. Open in a new tab Time evolution of structural parameters during the 500 ns MD simulation showing ( A ) RMSD, ( B ) RMSF, ( C ) radius of gyration, and ( D ) solvent-accessible surface area. Hydrogen bond dynamics : Throughout the entire 500 nanosecond simulation (Fig. 9 A), the hydrogen bond profile of the VEGFR2 and lead compound complex showed great stability. The intramolecular hydrogen bonds of the protein fluctuated in frequency between 600 and 780 and congealed in averages of 680 to 720, in the manner indicative of a steady protein structure and composition. Intermolecular hydrogen bonds between ligand and VEGFR2 showed slight variation ranging between 8 and 22 with a running mean of around 12–15. These observations imply the presence of strong and persistent ligand-receptor interactions that are sufficient to firmly anchor the compound in the binding pocket. The lack of extended reductions in intermolecular hydrogen bonding further supports a conclusion that the lead compound did not dissociate or drift very far from Pocket 1 over the period of the simulation. Fig. 9. Open in a new tab ( A ) Hydrogen bond dynamics, ( B ) internal energy components, ( C ) dynamic cross-correlation matrix, and ( D ) principal component analysis 3D projection for the VEGFR2–lead complex. Internal energy analysis : There were consistently favorable interaction energy components between VEGFR2 and the lead compound (Fig. 9 B). The van der Waals contribution ranged from − 55 to -70 kcal/mol, with a mean value of ~ -60 kcal/mol, indicating the importance of hydrophobic packing and shape complementarity in the stable complex. Electrostatic interactions were from − 18 to -30 kcal/mol, with mean around − 24 kcal/mol, denoting stable polar contacts and hydrogen bonds during the course of the trajectory. There was no long-term drift or destabilization of either component. The stability of both electrostatic and van der Waals energies taken together proves the maintenance of robust and energetically favorable contacts of the ligand with VEGFR2 at both the start and at the end of the simulation. Dynamic cross-correlation matrix (DCCM) : The DCCM heatmap showed a well spread of dynamic coupling over the VEGFR2 residues (Fig. 9 C). Strong positive correlations approaching + 1.0 were observed along the main diagonal as expected concerted motions of sequential residues in the protein backbone. Several off-diagonal correlated regions appeared as clusters of green and yellow patches, indicating synchronized domain-level motions needed to maintain the structural flexibility of the protein. There were regions of low anticorrelation (peak average of minus 0.2 to minus 0.4) scattered across the matrix that reflected natural hinge motions at loop and domain boundaries. Overall, the DCCM pattern shows that during ligand binding VEGFR2 preserved stable collective motions and did not undergo large-scale domain rearrangements supporting the integrity of the ligand-receptor interactions that were revealed in docking and molecular dynamics analyses. The absence of strong long-range anticorrelated motions further indicates that ligand binding does not induce disruptive conformational rearrangements, supporting stable allosteric communication within the kinase domain. Principal component analysis (PCA) : PCA results showed a very clear evolution from an initial flexible state to a more restricted and stable conformational ensemble during the simulation (Fig. 9 D). Early trajectory frames (0-150 ns) were very spread out in PC1 and PC2, indicating more conformational freedom upon initial equilibration. After about 200 ns, the points started to aggregate more closely with the densest grouping occurring at 300 to 500 ns, suggesting the protein-ligand complex was in a stable energy basin. This convergence towards a compact cluster in PCA space reflects the stabilization of global motions of VEGFR2, and confirms that the system reached long time scale equilibrium. Such behavior is consistent with RMSD, rGyr and energy profiles, which all showed similar stabilization over similar timescales. Principal component analysis was applied to the molecular dynamics trajectories to capture the dominant collective motions of the VEGFR2–derivative 7 complex. The first few principal components accounted for the majority of conformational variance, indicating that large-scale motions were restricted to a limited subspace. Projection of the trajectory along the principal components revealed confined conformational sampling, suggesting structural stability of the complex during the simulation period. The absence of abrupt transitions or highly dispersed conformations supports the persistence of a stable binding mode under dynamic conditions. MM/GBSA and MM/PBSA binding free energy analysis: Binding free energy estimations by MM/GBSA showed that there was a very favorable interaction between the lead compound and VEGFR2. The calculated value of the total free energy change, Rosetta Fold Analysis (In total MM/GBSA, ΔG total = -55.13 kcal/mol), in which the ligand forms a strongly stabilized complex driven by combination of van der Waals and electrostatic contributions (Table 9 ). This value is in accord with the stable interaction patterns seen throughout the molecular-dynamics trajectory and indicates a strongly bound ligand with few energetic penalties. Table 9. MM/GBSA binding free energy components calculated from 5000 frames of the MD simulation. Energy component Average Std. dev. Std. err. of mean VDWAALS -59.1996 2.8740 0.8666 EEL -41.4439 3.2560 0.9817 EGB 53.9068 2.2036 0.6644 ESURF -8.3934 0.1031 0.0311 DELTA G gas -100.6436 4.0837 1.2313 DELTA G solv 45.5135 2.1910 0.6606 DELTA TOTAL -55.1301 3.8932 1.1739 Open in a new tab The MM/PBSA evaluation brought a complementary point of view to get a ΔG total of -12.03 kcal/mol for the VEGFR2 - lead compound complex (Table 10 ). Although this value is less negative than the MM/GBSA estimate it is still favorable and confirms that the ligand stays in a stable association with the binding site also when solvation effects are modelled by the Poisson Boltzmann framework. Together, the free energy results support the conclusion that the lead compound possesses the strong thermodynamic affinity for VEGFR2 and behaves as a stable inhibitor candidate under dynamic physiological conditions. Table 10. MM/PBSA binding free energy components calculated from 5000 frames of the MD simulation. Energy component Average Std. dev. Std. err. of mean VDWAALS -59.1996 2.8740 0.8666 EEL -41.4439 3.2560 0.9817 EPB 59.0801 2.2896 0.6903 ENPOLAR -43.5078 0.4668 0.1407 EDISPER 73.0378 0.7798 0.2351 DELTA G gas -100.6436 4.0837 1.2313 DELTA G solv 88.6101 2.4842 0.7490 DELTA TOTAL -12.0334 3.7053 1.1172 Open in a new tab The combined docking, interaction, and molecular dynamics analyses reproducibly select AI -Fragmented Derivative 7 (Grow) as the most potent VEGFR2 inhibitor out of the generated analogues. During the docking stage, Derivative 7 showed the highest binding free energy value of the series (ΔG = -11.7 kcal mol-1) over both reference agent Sorafenib and the rest of the analogues. Its binding conformation is closely packed into the Pocket1, establishing a dense web of hydrophobic contacts with key amino acid residues Val848, Ala866, Leu889, Val899, Val916, Leu1035 and Phe1047, as well as a number of hydrogen bond interactions between the side chain of the glutamate terminal group of glutamic acid residue 885, the amino acid valve, and the amino acid residues cysteine 919; besides, the active binding sites contain the amino acids glycine 922 and aspartic acid 10 This very geometrically favorable pattern of interaction implies considerable ligand-receptor complementarity. The molecular dynamics simulations, which were run over an extended period of time, further confirm the favorable docking conformation is not a transient artifact but is robust under near- physiologic conditions. Root mean square deviation (RMSD) values of the protein, its active site and the ligand converge quickly with very little fluctuation and the result is very similar, while root mean square fluctuation (RMSF) analysis shows little flexibility of binding site residues indicating structural organization of the binding site aimed at stabilization around the ligand. The radius of gyration and solvent accessible surface area (SASA) profiles are maintained at nearly constant values following this initial equilibration phase, which thus reveals conservation of the overall protein fold and solvent exposure, and indicates that engagement of the ligand does not induce destabilizing expansion or collapse. Hydrogen-bond analyses of the entire 500-nanosecond trajectory reveal a robust ensemble of intermolecular hydrogen bonds between Derivative 7 and VEGFR2 which occurs concurrently with a stable intramolecular hydrogen-bond network within the protein, implying formation of a compact, strongly associated complex. Consistently favorable van der Waals and electrostatic interaction energies as well as estimates of a free energy change (Δtotal) from MM/GBSA calculations and MM/PBSA calculations support thermodynamically stable binding. The dynamic cross correlation matrix (DCCM) shows correlated motions organized rather than disruptive domain rearrangements and principal component analysis (PCA) shows a clear convergence towards a stable conformational basin after initial equilibration. Taken together, the combination of high docking affinity, favorable hydrogen bonding and hydrophobic interaction network, the excellent structural and energetic stability in simulation, and the favorable predictions for the binding free energy strongly support AI-Fragmented Derivative 7 (Grow) as the most promising drug candidate in the library derived from Sorafenib for targeting VEGFR2. Discussion There is an ongoing need for VEGFR2-directed inhibitors with improved safety and drug-like properties, since many clinically used VEGFR2 inhibitors (including multi-kinase agents such as Sorafenib and Regorafenib) are associated with dose-limiting toxicities 8 . Sorafenib is a known multi-kinase inhibitor targeting VEGFR1–3, PDGFR, c-Kit, and RAF kinases, a property that contributes to its clinical efficacy but is also associated with dose-limiting adverse effects 39 . This toxicity profile combined with lack of selectivity makes use of such agents unsuitable for pregnancy or placental disorders. In the present study, the rationale for designing Sorafenib derivatives was not to demonstrate improved kinase selectivity but to computationally optimize VEGFR2 binding characteristics while preserving drug-like properties. Accordingly, the proposed derivative represents an in silico optimized candidate with enhanced target engagement toward VEGFR2, while improvements in kinase selectivity remain to be experimentally and computationally evaluated in future studies. Although VEGF–VEGFR2 signaling is essential for physiological placental vascular development, persistent or dysregulated activation of this pathway has been implicated in pathological angiogenesis and endothelial instability. In such contexts, controlled or partial inhibition of VEGFR2 has been proposed as a means to normalize aberrant angiogenic signaling rather than suppress vascular formation. Accordingly, the present study investigates VEGFR2 inhibition as a modulatory strategy aimed at attenuating pathological angiogenic responses. Consistent with recent fragment-based computational studies on VEGFR2 inhibitors, the present work emphasizes rational optimization of binding features and interaction stability as a preliminary step toward improving target engagement while acknowledging that experimental validation is required to assess selectivity and toxicity profiles 11 . The lead sorafenib derivative (Derivative 7 (Grow)) showed a significantly increased docking affinity for VEGFR2, with a binding energy of -11.7 kcal/mol compared to -10.2 kcal/mol for sorafenib. This in silico affinity gain is consistent with empirical observations for the fluorinated sorafenib analogue regorafenib which has a better potency as a VEGFR2 inhibitor 40 . Docking pose showed that Derivative 7 (Grow) maintains the same characteristic interactions of type-2 VEGFR2 kinase inhibitors: hydrogen bonds with the hinge residue, Cys919, and with the DFG (Asp1046) motif, and it occupies a neighboring hydrophobic pocket near the ATP site. Sorafenib itself binds this way, and the preservation of these contacts alongside other additional hydrophobic contacts (with Val848, Leu889, Val916, Phe1047) and hydrogen bonds with the side chains of the residues at the same site, for example with the side chain of the side chain of the residue at the same site) contributes to its higher binding score. Now the introduction of functional groups, such as amides or esters in strategic positions, could create more hydrogen bonds to the solution to the surface protein (Cys919) thereby improving the affinity 8 . In the present study, fragmentation-based optimization enhanced interaction quality within the known VEGFR2 binding pocket; however, no novel sub-pocket or alternative binding mode was identified. Likewise, electron withdrawing substituents on the scaffold have previously been found to enhance ligand-receptor interactions, providing further evidence to support the observed enhancements over the parent drug. Accordingly, compounds exhibiting slightly less favorable docking scores but better stability, interaction persistence, and pharmacokinetic profiles were considered more promising leads than those selected solely on the basis of binding affinity. MD and MM/GBSA/MM/PBSA analyses were employed as confirmatory tools to evaluate the dynamic stability and energetic favorability of the selected lead rather than as a comparative screening method across all derivatives. Long timescale molecular dynamics simulations (500 ns) revealed that the lead compound forms an incredibly stable complex with VEGFR2. Protein backbone RMSD stayed within 0.13–0.18 nm and the ligand RMSD showed fluctuation only in the range of 0.05–0.1 nm, indicating that the initial docked conformation is retained without significant deviation. This stability is similar to earlier MD studies of high-affinity inhibitors of VEGFR2, for example a 100- ns simulation of a potent nicotinamide-based inhibitor that kept a backbone RMSD of ~ 2.5 Å 41 . In our system, the equilibrium lasted for a longer period of time (500 ns), highlighting the potential of the derivative to “lock” into the binding pocket. The radius of gyration was constant and there was normal fluctuation in solvent accessible surface area, which proved that the binding of the ligand does not lead to an unfolding or collapse of the receptor. Throughout the trajectory, a stable network of hydrogen bonds (on average, ~ 12–15 intermolecular H-bonds) showed a correlation with high binding stability, in line with earlier reports that persistent hydrogen bonds improve inhibitor stability 41 . PCA provided complementary insight into the dynamic behavior of the complex by revealing restricted collective motions and limited exploration of the conformational space sampled during the simulation. Although free energy landscape analysis can further delineate energetic basins, the present PCA-based evaluation was intended as a qualitative descriptor of dynamic stability rather than an energetic analysis. Such interpretation of principal motions is widely applied in MD-based lead validation studies and indicates preserved intramolecular communication and conformational coherence of VEGFR2 upon ligand binding. A recent comparative simulation further proved these observations: a 500 ns MD simulation of a designed VEGFR2 inhibitor revealed better receptor stabilization than sorafenib; this was similar to the trend for derivative 7 42 . Further potent activity of the derivative can be rationalized through pharmacophore analysis. The lead compound preserves the critical pharmacophoric features of sorafenib, including two aromatic rings for pi-pi stacking and hydrophobic interactions, several hydrogen-bonding donors and acceptors, and hydrophobic tails, and makes subtle improvements. Notably, derivative 7 contains an additional hydrogen-bond acceptor compared to sorafenib (four accepted hydrogen bonds compared to three) while keeping the donor-acceptor number of hydrogen bonds constant and is likely responsible for the extra hydrogen-bonds observed (e.g., interactions with Glu885 or Asp1046). This result is in agreement with known structure-activity relationships for VEGFR2 inhibitors, where polar moieties that interact with hinge or DFG residues increase affinity 8 . The derivative fits the general formula of the chemotherapeutic agent for treating various tumors, VEGFR2 inhibitors: a heterocyclic nucleus filling the ATP-binding site, the central aryl nucleus, and a hydrophobic end extending into the allosteric back pocket with a length of the linker (3–5 bonds) as that of sorafenib, which protects the complementarity in space. Density functional theory calculations gave further insight into the electronic profile of the compound. The HOMO-LUMO gap of lead derivative (0.154 a.u) is slightly less than that of sorafenib (0.162 a.u), which means greater electronic polarizability. A smaller gap generally allows for easier transfer of charge and potentially leads to increased reactivity in intermolecular interactions 43 . Accordingly, the derivative’s electron density is more flexible upon binding, which may increase electrostatic attractions with VEGFR2. This is supported by a molecular electrostatic potential map which highlights electron rich regions complementary to the active site (similar to those seen in hydrogen bonds/salt bridges during MD). This fine tuning, as electronically explained, is in line with precedence in the design of VEGFR2 inhibitors, wherein electron-donating or withdrawing substituents can modulate potency, reinforcing the idea that the enhanced activity of derivative 7 is due to a balanced electronic structure, neither over- nor under-reacting 8 . Pharmacophore modeling and HOMO–LUMO gap analysis were applied to the prioritized lead to obtain an in-depth understanding of its binding features and electronic properties, rather than to comparatively rank structurally similar screening candidates. Correlation of DFT outcomes with docking and molecular dynamics results further supports the inhibitory potential of the selected lead. Electron-rich regions identified in the HOMO and MEP analyses align with key hydrogen bonding and polar interactions observed with critical VEGFR2 active-site residues, while electron-deficient regions correspond to areas involved in stabilizing van der Waals and electrostatic interactions 44 . The moderate HOMO–LUMO gap is consistent with the stable binding behavior observed during MD simulations and favorable MM/GBSA binding free energy estimates 45 . Together, these findings suggest that the electronic characteristics of derivative 7 contribute to its binding stability and potential VEGFR2 inhibitory activity. Critical evaluation of pharmacokinetic and toxicity profiles further highlights the advantages the derivative. In silico ADMET predictions suggest that Derivative 7 (Grow) meets criteria for drug-like compounds (there are no Lipinski violations) even though it has a slightly higher molecular weight. Its predicted lipophilicity (logP ~ 3.7) is slightly lower than that of sorafenib (logP ~ 4.1) and its higher topological polar surface area indicates its improved solubility and moderate distribution. Accordingly, the derivative has a somewhat improved aqueous solubility (around − 4.8 log S vs. -5.1 for sorafenib) and similar intestinal permeability. Both compounds are predicted to have poor gastrointestinal absorption and negligible blood-brain barrier-followed by -barrier penetration; the latter is preferred for a VEGFR2 inhibitor for peripheral use, as it reduces central nervous system side effects. Cytochrome P450 inhibition profiles are similar, which suggests similar metabolic pathways. Importantly, toxicity predictions show that sorafenib is indicated for hepatotoxicity (drug version liver injury), whereas derivative 7 is safety indicated in the DILI version of the model. This differential indicates a reduced tendency for liver damage, which addresses an important limitation of sorafenib, which boasts the highest reported incidence of serious liver injury of the VEGFR-targeted TKIs 46 . Derivative 7 (Grow) does not have any alerts for mutagenicity or cardiotoxicity, like sorafenib across other end points. These results are supported by clinical observations of other selective VEGFR2-inhibitors. Apatinib, a newer, smaller molecule, selective blocker of VEGFR2, also shows meaningful anti-angiogenic efficacy in oncology with a manageable toxicity profile 40 . Thus, targeting VEGFR2 and minimizing off target kinase interactions seems to translate to side effects. Derivative 7 (Grow) is consistent with the paradigm of computationally optimizing VEGFR2 binding characteristics while maintaining drug-like properties. If predicted liver safety and overall tolerability are confirmed experimentally, it would address a major limiting factor of current anti-VEGF therapies, which provides increased suitable for more sensitive populations. Importantly, the present study does not evaluate maternal–fetal safety, placental transfer, or pregnancy-specific pharmacology. Therefore, no conclusions can be drawn regarding suitability of the proposed derivative in pregnancy-associated conditions. Future translational studies would require rigorous maternal fetal safety testing and placental permeability assessment prior to any consideration of pregnancy related applications. 47 . The resulting improved therapeutic index that the ADMET analysis indicates is thus a promising advance. Comparative analysis with Sorafenib and Regorafenib suggests several potential advantages of Derivative 7 at the computational level. First, the derivative retained the core pharmacophoric features required for type-II VEGFR2 binding while introducing structural modifications that improved predicted binding complementarity within the VEGFR2 ATP-binding pocket. Second, the compound demonstrated improved docking and MM/GBSA binding energetics compared to Sorafenib, suggesting stronger target engagement within the VEGFR2 kinase domain. Third, molecular dynamics analysis indicated stable complex formation and persistent interaction profiles over the simulation timeframe. Finally, in silico ADMET prediction suggested improved aqueous solubility and a reduced hepatotoxicity flag compared to Sorafenib. Importantly, this study did not perform kinome-wide off-target profiling; therefore, any improvement in kinase selectivity over Sorafenib remains to be determined through future multi-target computational screening and experimental validation. Despite promising in silico results, this study is purely computational. While docking, MD, DFT and ADMET predictions are powerful tools, they have certain intrinsic limitations. Biological activity must be confirmed by in vitro enzyme assays and cellular studies; a compound that has a favorable docking interaction in silico may not inhibit VEGFR2 in a cell environment due to models that do not represent protein dynamics or permeability. The next step, therefore, is the biosynthesis of the top performing derivative and the experimental evaluation of its activity as a VEGFR2 inhibitor, in comparison to sorafenib and regorafenib. This in silico study does not propose indiscriminate inhibition of VEGFR2 during normal pregnancy. Instead, it evaluates VEGFR2 modulation as a potential therapeutic strategy in conditions associated with pathological or dysregulated angiogenic signaling, pending experimental and clinical validation. Additionally, although this research focused on VEGFR2, selectivity against other kinases (e.g. VEGFR1, VEGFR3, PDGFR, c-Kit) will need to be quantified to ensure that no significant off-target inhibition compromises the predicted safety advantage. Pharmacokinetic predictions require empirical validation by metabolic stability studies and toxicity studies (e.g. hepatic cytotoxicity assays, hERG channel assays). Ligand efficiency analysis can provide additional insight during early-stage screening, comprehensive lead optimization requires multi-dimensional evaluation beyond size-normalized docking metrics. While this 500ns MD simulation has been useful and informative, it cannot represent all in vivo behavior; improved sampling or longer simulations could well be justified to include the possibility of slow developing changes in conformation. This study does not provide a comparative kinome-wide selectivity analysis; therefore, potential improvements in target specificity of the proposed derivative over Sorafenib remain to be determined through future multi-target computational and experimental profiling. For the application of placental disorders additional considerations arise. Any candidate drug for patients with a pregnancy must be tested for placental transfer and fetal safety in animal models. Subsequent studies should involve in vivo models of placental angiogenesis disorders (such as pregnant rodents with induced preeclampsia), determining if the inhibitor modulates angiogenesis of the placenta vessels without adverse maternal or fetal consequences. Formulation development may also be pursued, as a result of the increased polarity of the derivative, to ensure adequate bioavailability in vivo. Comparative molecular dynamics simulations of multiple derivatives were beyond the scope of this study and may be explored in future work to further refine lead prioritization. Finally, both the AI guided scaffold modification and multi parameter optimization strategy presented herein can be repetitively applied. Should it be that Derivative 7 (Grow) does not work optimally, there may be iterations that include metabolic stabilizers or potency-enhancing fragments. The present data pave the way for a new generation of VEGFR2 inhibitors based on a sorafenib-derived core; the upcoming experimental investigations will archive this in silico success into real therapeutic potential of aberrant placental angiogenesis. Conclusion This study identified and characterized a VEGFR2-directed lead candidate (AI-Fragmented Derivative 7, Grow) derived from the Sorafenib scaffold through an AI-guided fragmentation and structure-based computational optimization workflow. Among the twenty generated derivatives, Derivative 7 exhibited improved predicted binding performance toward the VEGFR2 kinase domain compared to the parent compound, supported by docking analysis and preservation of key type-II kinase inhibitor interactions within the ATP-binding region. The stability and dynamic persistence of the VEGFR2–Derivative 7 complex were further validated through 500 ns molecular dynamics simulations, demonstrating stable RMSD behavior, persistent hydrogen bonding patterns, and consistent compactness and solvent exposure profiles. Binding free energy estimations using MM/GBSA and MM/PBSA calculations provided additional thermodynamic support for favorable ligand–receptor association. In parallel, electronic structure analysis by DFT suggested improved reactivity descriptors, including a narrower HOMO–LUMO energy gap relative to Sorafenib, which may contribute to enhanced binding compatibility. Pharmacokinetic profiling further indicated that Derivative 7 possesses improved predicted aqueous solubility and reduced hepatotoxicity risk compared to Sorafenib, supporting its potential as a more drug-like VEGFR2-targeted candidate at the in-silico level. Importantly, this work is purely computational and does not provide experimental confirmation of inhibitory potency, selectivity across the kinome, or biological efficacy. Therefore, the proposed lead compound should be considered a computationally prioritized candidate for future validation through biochemical VEGFR2 inhibition assays, cellular angiogenesis models, and broader off-target kinase profiling. Overall, the present findings demonstrate the applicability of AI-assisted fragmentation strategies integrated with molecular docking, molecular dynamics, binding free energy analysis, ADMET prediction, and DFT descriptors for rational optimization of VEGFR2 inhibitors relevant to angiogenesis-associated pathologies. These findings are particularly relevant to pathological conditions characterized by aberrant or excessive VEGFR2-driven angiogenesis, including solid tumors, ocular neovascular disorders, inflammatory angiogenesis, and fibrotic remodeling. By providing a computationally optimized VEGFR2-binding scaffold, this study offers a foundation for further preclinical development in established VEGFR2-associated disease settings. Acknowledgements We would like to acknowledge that no individuals or organizations were involved in the development, funding, or support of this work. This research was solely conducted by the authors, and no external contributions, financial or otherwise, were received or utilized in the execution of this study. Author contributions Dr. Deniz Inan contributed to the conceptualization, overall supervision of the study, and critical review and editing of the manuscript. Dr. Sinan Karageçili was involved in methodology development, validation of computational approaches, and manuscript review. Nouman Ali carried out data curation, formal analysis, computational investigations, molecular modeling, and visualization, and was primarily responsible for drafting the original manuscript as well as subsequent revisions. Funding This research was conducted without external funding. The authors would like to clarify that no financial support or grants from any organization, institution, or individual were received for the completion of this review. The research was carried out as part of the author’s academic and professional responsibilities, and they did not rely on any external sources of funding for the design, execution, or publication of this work. Data availability All data generated or analyzed during this study are included in this published article. The datasets supporting the conclusions of this study including ligand structures (SDF and PDB files), receptor structures, active site prediction, DFT, molecular docking results, and molecular dynamics simulation outputs are available in a publicly accessible Zenodo repository: https:/doi.org/10.5281/zenodo.17701237. Declarations Competing interests The authors declare no competing interests. This research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Contributor Information Deniz Inan, Email: [email protected]. Nouman Ali, Email: [email protected]. References 1. Martowicz, A. et al. Endothelial β-Catenin signaling supports postnatal brain and retinal angiogenesis by promoting sprouting, tip cell formation, and VEGFR (vascular endothelial growth factor receptor) 2 expression. Arterioscler. Thromb. Vasc. Biol. 39 , 2273–2288. 10.1161/ATVBAHA.119.312749 (2019). [ DOI ] [ PubMed ] [ Google Scholar ] 2. Dan, H. et al. CM082, a novel VEGF receptor tyrosine kinase inhibitor, can inhibit angiogenesis in vitro and in vivo. Microvasc. Res. 136 , 104146. 10.1016/j.mvr.2021.104146 (2021). [ DOI ] [ PubMed ] [ Google Scholar ] 3. Wilhelm, S. M. et al. Preclinical overview of sorafenib, a multikinase inhibitor that targets both Raf and VEGF and PDGF receptor tyrosine kinase signaling. Mol. Cancer Ther. 7 , 3129–3140. 10.1158/1535-7163.MCT-08-0013 (2008). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 4. Meyer, M. et al. A novel vascular endothelial growth factor encoded by Orf virus, VEGF-E, mediates angiogenesis via signalling through VEGFR‐2 (KDR) but not VEGFR‐1 (Flt‐1) receptor tyrosine kinases. EMBO J. 18 , 363–374. 10.1093/emboj/18.2.363 (1999). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 5. Papapanagiotou, A. et al. The Role of Angiogenetic Factors in Preeclampsia. Int. J. Mol. Sci. 26 , 10431 (2025). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 6. Adnane, L., Trail, P. A., Taylor, I. & Wilhelm, S. M. in Methods in Enzymology Vol. 407 597–612Academic Press, (2006). [ DOI ] [ PubMed ] 7. El-Hazek, R. M. M., Zaher, N. H., El-Gazzar, M. G. M., Fadel, N. A. & El-Sabbagh, W. A. Novel VEGFR2 inhibitors with thiazoloquinoxaline scaffold targeting hepatocellular carcinoma with lower cardiotoxic impact. Sci. Rep. 13 , 13907. 10.1038/s41598-023-40832-z (2023). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 8. Chaudhari, P. J., Nemade, A. R. & Shirkhedkar, A. A. Recent updates on potential of VEGFR-2 small-molecule inhibitors as anticancer agents. RSC Adv. 14 , 33384–33417. 10.1039/D4RA05244G (2024). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 9. Platzer, G., Mayer, M., McConnell, D. B. & Konrat, R. NMR-driven structure-based drug discovery by unveiling molecular interactions. Commun. Chem. 8 , 167. 10.1038/s42004-025-01542-x (2025). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 10. Buzatu, I. M. et al. A Review of FDA-Approved Multi-Target Angiogenesis Drugs for Brain Tumor Therapy. Int. J. Mol. Sci. 26 , 2192 (2025). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 11. Banerjee, S. et al. Fragment-based investigation of thiourea derivatives as VEGFR-2 inhibitors: A cross-validated approach of ligand-based and structure-based molecular modeling studies. J. Biomol. Struct. Dynamics . 42 , 1047–1063. 10.1080/07391102.2023.2198039 (2024). [ DOI ] [ PubMed ] [ Google Scholar ] 12. Li, Q. Application of Fragment-Based Drug Discovery to Versatile Targets. Front. Mol. Biosci. 7–2020. 10.3389/fmolb.2020.00180 (2020). [ DOI ] [ PMC free article ] [ PubMed ] 13. Velankar, S., Burley, S. K., Kurisu, G., Hoch, J. C. & Markley, J. L. The protein data bank archive. Structural Proteomics: High-Throughput Methods , 3–21 (2021). [ DOI ] [ PubMed ] 14. Hussain, M. et al. Computational modeling of cyclotides as antimicrobial agents against Neisseria gonorrhoeae PorB porin protein: Integration of docking, immune, and molecular dynamics simulations. Front. Chem. 12 , 1493165 (2024). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 15. Naveed, M. et al. An in silico approach uncovering the competency of oncolytic human adenovirus 52 for targeted breast cancer virotherapy. Sci. Rep. 14 , 26405 (2024). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 16. Colovos, C. & Yeates, T. O. Verification of protein structures: Patterns of nonbonded atomic interactions. Protein Sci. 2 , 1511–1519 (1993). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 17. Naveed, M. et al. Assessment of Melia azedarach plant extracts activity against hypothetical protein of mycobacterium tuberculosis via GC-MS analysis and in-silico approaches. J. Comput. Biophys. Chem. 23 , 299–320 (2024). [ Google Scholar ] 18. Mehryary, F., Nastou, K., Ohta, T., Jensen, L. J. & Pyysalo, S. STRING-ing together protein complexes: Corpus and methods for extracting physical protein interactions from the biomedical literature. Bioinformatics 40 , btae552 (2024). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 19. Kim, S. et al. PubChem substance and compound databases. Nucleic Acids Res. 44 , D1202–D1213 (2016). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 20. Asrar, B., Ali, N., Ali, I. & Naveed, M. In silico investigation of ketamine and methylphenidate drug-drug conjugate for MDD and ADHD treatment using MD simulations and MMGBSA. Sci. Rep. 15 , 24565 (2025). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 21. Torrens-Fontanals, M., Tourlas, P., Doerr, S. & De Fabritiis, G. PlayMolecule Viewer: A toolkit for the visualization of molecules and other data. J. Chem. Inf. Model. 64 , 584–589 (2024). [ DOI ] [ PubMed ] [ Google Scholar ] 22. Ali, N., Asrar, B. & Ahmad, S. A. I. Fragment Based Optimization of Saffron and Chamomile Phytochemicals as Aryl Hydrocarbon Receptor Inhibitors for Dementia Therapy an Integrated Computational Approach. Computational Biology Chemistry , 108606 (2025). [ DOI ] [ PubMed ] 23. Polák, L. et al. PrankWeb 4: A modular web server for protein–ligand binding site prediction and downstream analysis. Nucleic Acids Research , gkaf421 (2025). [ DOI ] [ PMC free article ] [ PubMed ] 24. Jejurikar, B. L. & Rohane, S. H. Drug designing in discovery studio. (2021). 25. Eastman, P. et al. OpenMM 7: Rapid development of high performance algorithms for molecular dynamics. PLoS Comput. Biol. 13 , e1005659. 10.1371/journal.pcbi.1005659 (2017). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 26. Guex, N., Peitsch, M. C. & Schwede, T. Automated comparative protein structure modeling with SWISS-MODEL and Swiss‐PdbViewer: A historical perspective. Electrophoresis 30 , S162–S173 (2009). [ DOI ] [ PubMed ] [ Google Scholar ] 27. Huey, R., Morris, G. M. & Forli, S. Using AutoDock 4 and AutoDock vina with AutoDockTools: A tutorial. Scripps Res. Inst. Mol. Graphics Lab. 10550 , 1000 (2012). [ Google Scholar ] 28. Naveed, M. et al. The natural breakthrough: Phytochemicals as potent therapeutic agents against spinocerebellar ataxia type 3. Sci. Rep. 14 , 1529 (2024). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 29. Pawar, R. P. & Rohane, S. H. Role of autodock vina in PyRx molecular docking. (2021). 30. Naveed, M. et al. Exploring the therapeutic potential of Thymus vulgaris ethanol extract: A computational screening for antimicrobial compounds against COVID-19 induced mucormycosis. Sci. Rep. 15 , 15906 (2025). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 31. Schake, P., Bolz, S. N., Linnemann, K. & Schroeder, M. PLIP 2025: Introducing protein–protein interactions to the protein–ligand interaction profiler. Nucleic Acids Res. 53 , W463–W465. 10.1093/nar/gkaf361 (2025). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 32. Rahman, S. U. et al. Innovative Phytochemicals Based In-Silico Drug Design and Molecular Dynamics Simulation Targeting Norovirus Induced Gastroenteritis. Journal Mol. Structure , 141733 (2025). 33. Daina, A., Michielin, O. & Zoete, V. SwissADME: A free web tool to evaluate pharmacokinetics, drug-likeness and medicinal chemistry friendliness of small molecules. Sci. Rep. 7 , 42717 (2017). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 34. Myung, Y., de Sá, A. G. & Ascher, D. B. Deep-PK: Deep learning for small molecule pharmacokinetic and toxicity prediction. Nucleic Acids Research , gkae254 (2024). [ DOI ] [ PMC free article ] [ PubMed ] 35. Koes, D. R. The Pharmit backend: A computer systems approach to enabling interactive online drug discovery. IBM J. Res. Dev. 62, 3 : (1–3), 6 (2018). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 36. Ali, N., Akbar, R., Saleem, A., Ali, A. & Ali A. AI based natural inhibitor targeting RPS20 for colorectal cancer treatment using integrated computational approaches. Sci. Rep. 15 , 24906 (2025). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 37. Kashefolgheta, S., Wang, S., Acree, W. E. & Hünenberger, P. H. Evaluation of nine condensed-phase force fields of the GROMOS, CHARMM, OPLS, AMBER, and OpenFF families against experimental cross-solvation free energies. Phys. Chem. Chem. Phys. 23 , 13055–13074 (2021). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 38. Zaheer, M., Ali, N., Javed, H., Munir, R. & Jamil, N. Uncovering the impact of SARS-CoV2 spike protein variants on human receptors: A molecular dynamics docking and simulation approach. J. Infect. Public Health . 16 , 1544–1555 (2023). [ DOI ] [ PubMed ] [ Google Scholar ] 39. Pang, Y. et al. Adverse events of sorafenib in hepatocellular carcinoma treatment. Am. J. Cancer Res. 12 , 2770–2782 (2022). [ PMC free article ] [ PubMed ] [ Google Scholar ] 40. Mou, L. et al. Improving Outcomes of Tyrosine Kinase Inhibitors in Hepatocellular Carcinoma: New Data and Ongoing Trials. Front. Oncol. 11–2021 10.3389/fonc.2021.752725 (2021). [ DOI ] [ PMC free article ] [ PubMed ] 41. Elgammal, W. E. et al. New nicotinamide–thiadiazol hybrids as VEGFR-2 inhibitors for breast cancer therapy: Design, synthesis and in silico and in vitro evaluation. RSC Adv. 15 , 14477–14498 (2025). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 42. Wasai, S. et al. Unveiling Novel Vegfr-2 Inhibitors: Qsar Modeling, Molecular Docking, Md and Mm-Gbsa Calculations, Pharmacokinetics Profiling and Dft Studies. Molecular Docking, Md and Mm-Gbsa Calculations, Pharmacokinetics Profiling and Dft Studies . 43. Mahal, A. et al. Molecular docking, drug-likeness and DFT study of some modified tetrahydrocurcumins as potential anticancer agents. Saudi Pharm. J. 32 , 101889. 10.1016/j.jsps.2023.101889 (2024). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 44. Banerjee, S., Jana, S., Jha, T., Ghosh, B. & Adhikari, N. An assessment of crucial structural contributors of HDAC6 inhibitors through fragment-based non-linear pattern recognition and molecular dynamics simulation approaches. Comput. Biol. Chem. 110 , 108051. 10.1016/j.compbiolchem.2024.108051 (2024). [ DOI ] [ PubMed ] [ Google Scholar ] 45. Baidya, S. K., Banerjee, S., Ghosh, B., Jha, T. & Adhikari, N. Pinpointing prime structural attributes of potential MMP-2 inhibitors comprising alkyl/arylsulfonyl pyrrolidine scaffold: a ligand-based molecular modelling approach validated by molecular dynamics simulation analysis. SAR QSAR Environ. Res. 35 , 665–692. 10.1080/1062936X.2024.2389822 (2024). [ DOI ] [ PubMed ] [ Google Scholar ] 46. Wang, X., Chen, R., Liu, J., Wang, E. & Luo, H. Liver injury related to vascular endothelial growth factor tyrosine kinase inhibitors: A pharmacovigilance analysis of the USA FDA adverse event reporting system (FAERS) database. Expert Opin. Drug Saf. 1–9. 10.1080/14740338.2025.2460449 (2025). [ DOI ] [ PubMed ] 47. Elkaeed, E. B. et al. Discovery of New VEGFR-2 Inhibitors: Design, Synthesis, Anti-Proliferative Evaluation, Docking, and MD Simulation Studies. Molecules 27 10.3390/molecules27196203 (2022). [ DOI ] [ PMC free article ] [ PubMed ] Associated Data This section collects any data citations, data availability statements, or supplementary materials included in this article. Data Availability Statement All data generated or analyzed during this study are included in this published article. The datasets supporting the conclusions of this study including ligand structures (SDF and PDB files), receptor structures, active site prediction, DFT, molecular docking results, and molecular dynamics simulation outputs are available in a publicly accessible Zenodo repository: https:/doi.org/10.5281/zenodo.17701237. Articles from Scientific Reports are provided here courtesy of Nature Publishing Group ACTIONS View on publisher site PDF (3.5 MB) Cite Collections Permalink PERMALINK Copy RESOURCES Similar articles Cited by other articles Links to NCBI Databases Cite Copy Download .nbib .nbib Format: AMA APA MLA NLM Add to Collections Create a new collection Add to an existing collection Name your collection * Choose a collection Unable to load your collection due to an error Please try again Add Cancel Follow NCBI NCBI on X (formerly known as Twitter) NCBI on Facebook NCBI on LinkedIn NCBI on GitHub NCBI RSS feed Connect with NLM NLM on X (formerly known as Twitter) NLM on Facebook NLM on YouTube National Library of Medicine 8600 Rockville Pike Bethesda, MD 20894 Web Policies FOIA HHS Vulnerability Disclosure Help Accessibility Careers NLM NIH HHS USA.gov Back to Top