CytoLNCpred: A computational method to predict the probability of lncRNA localizing to cytoplasm

CytoLNCpred: A computational method to predict the probability of lncRNA localizing to cytoplasm | Zenodo Skip to main Communities My dashboard Log in Sign up Zenodo is currently experiencing slowness and intermittent outages due to heavy automated traffic from bots and AI crawlers. We are aware of the problem, and our team is focused on stabilizing the service. Thank you for your patience. raghavagps Published April 29, 2026 | Version v1.0 Software Open CytoLNCpred: A computational method to predict the probability of lncRNA localizing to cytoplasm Authors/Creators Choudhury, Shubham (Researcher) 1 Raghava, Gajendra (Supervisor) 2 Show affiliations 1.

Indraprastha Institute of Information Technology 2. Indraprastha Institute of Information Technology Delhi Description Title: CytoLNCpred Dataset – Cell-line specific long non-coding RNAs (lncRNAs) with cytoplasm/nucleus localization annotations for 15 human cell lines Description: Project: CytoLNCpred – A computational method for predicting cytoplasm-associated long non-coding RNAs in 15 cell-lines Publication: Choudhury, S., Mehta, N.K., & Raghava, G.P.S. (2025). CytoLNCpred – a computational method for predicting cytoplasm associated long non-coding RNAs in 15 cell-lines. Frontiers in Bioinformatics , 5, 1585794. https://doi.org/10.3389/fbinf.2025.1585794 Overview: This repository accompanies the CytoLNCpred publication and provides the first cell-line-specific, non‑redundant dataset for predicting lncRNA subcellular localization (cytoplasm vs. nucleus) across 15 human cell lines. Unlike previous methods that included mRNA transcripts and excluded lncRNAs with borderline localization differences, CytoLNCpred uses a clean, properly filtered dataset based on the Cytoplasm to Nucleus Relative Concentration Index (CNRCI) from the LncATLAS database. This is critical because lncRNA function is largely determined by its subcellular location, and the same lncRNA can localize differently across cell lines. Content: The repository contains cell-line-specific datasets for 15 human cell lines, derived from LncATLAS (Mas‑Ponte et al., 2017) with sequences obtained from ENSEMBL (version 112). After quality filtering and redundancy reduction (MeshClust at 90% sequence identity; length ≤ 10,000 nucleotides), the final dataset comprises: Training set: Varies by cell line (total ~18,000–25,000 samples across all cell lines) Validation set: Varies by cell line (20% of data, held out from training) 15 cell lines: A549, GM12878, H1.hESC, HeLa.S3, HepG2, HT1080, HUVEC, IMR.90, K562, MCF.7, NCI.H460, NHEK, SK.MEL.5, SK.N.DZ , SK.N.SH Localization labeling (based on CNRCI): CNRCI = log₂(Cytoplasmic expression (FPKM) / Nuclear expression (FPKM)) Cytoplasm (positive class): CNRCI > 0 (higher expression in cytoplasm) Nucleus (negative class): CNRCI < 0 (higher expression in nucleus) Excluded: CNRCI = 0 (no differential localization) Key Findings from Analysis: Cell-line specific variability: Some lncRNAs (e.g., LINC00852, SNHG3) show dramatic localization shifts across cell lines – nuclear in some, cytoplasmic in others – highlighting the need for cell-line-specific models. LINC00852: Cytoplasmic in lung carcinoma (NCI.H460) but nuclear in most other cell lines, consistent with literature showing its cytoplasmic role in MAPK pathway activation in lung adenocarcinoma. SNHG3: Cytoplasmic in stem-like/proliferative cells (H1.hESC, HepG2, GM12878) but nuclear in differentiated/epithelial cells (MCF.7, NHEK), reflecting context-dependent ceRNA vs. transcriptional regulatory roles. Nucleotide preference: Cytosine-based k-mers are enriched in cytoplasm-localized lncRNAs; thymine is enriched in nuclear-localized lncRNAs. GO & KEGG Enrichment Analysis (RNAenrich): Cytoplasmic lncRNAs – Enriched in: response to interferon‑beta (GO:0035456), positive regulation of apoptosis (GO:0043065), RNA splicing (GO:0008380), Ferroptosis (hsa04216), Autophagy (hsa04140) – immune signaling, post‑transcriptional regulation, cellular stress. Nuclear lncRNAs – Enriched in: eukaryotic 48S preinitiation complex (GO:0033290), regulation of rRNA transcription (GO:1901836), MLL1/2 complex (GO:0044665), Sterol Biosynthesis (hsa00100), Nucleotide Excision Repair (hsa03420) – transcriptional regulation, chromatin remodeling, genome maintenance. Model Performance (Average across 15 cell lines, validation set): Approach Features Best ML Algorithm Avg. AUC Avg. MCC Avg. Accuracy Composition‑based 123 features (Nfeature) RF / XGB / SVC / GaussianNB 0.7049 0.1965 0.704 Correlation‑based 1,100 features (Nfeature) RF / AdaBoost / GaussianNB 0.7089 0.2133 0.694 DNABERT‑2 embeddings (pre‑trained) 768-dim (max pooling) MLP / SVC / XGB 0.665 0.149 0.704 DNABERT‑2 embeddings (fine‑tuned) 768-dim (max pooling) SVC / MLP / GaussianNB 0.660 0.174 0.701 Fine‑tuned DNABERT‑2 (end‑to‑end) Tokenized sequences Fine‑tuned DNABERT‑2 0.634 0.127 0.730 mRMR‑selected (Top 500) 500 features (combined) Best per cell line 0.6946 — — Best overall: Correlation‑based features + ML – average AUC = 0.7089 , MCC = 0.2133 across 15 cell lines. Correlation‑based features outperformed LLM‑based models (DNABERT‑2), demonstrating that carefully engineered physicochemical features capture lncRNA localization signals more effectively than general‑purpose language model embeddings for this task. Best performance per cell line (Correlation features): Cell Line Best Model AUC MCC Sensitivity Specificity HeLa.S3 GaussianNB 0.779 0.272 0.783 0.585 HT1080 Gaussian Process 0.738 0.359 0.583 0.769 SK.N.DZ SVC 0.739 0.327 0.424 0.865 HUVEC AdaBoost 0.757 0.243 0.302 0.897 A549 GradientBoosting 0.757 0.381 0.567 0.803 Benchmarking Comparison (on same validation dataset – average AUC): Method Average AUC (across shared cell lines) CytoLNCpred – Correlation features 0.709 CytoLNCpred – Composition features 0.705 TACOS (tree‑based, compositional features) 0.676 lncLocator 2.0 (deep learning, word embeddings) 0.523 CytoLNCpred substantially outperforms existing methods, particularly lncLocator 2.0 (which relies on noisy datasets containing mRNAs) and TACOS (which excludes lncRNAs with fold‑change < 2). mRNA localization prediction (cross‑applicability test): CytoLNCpred was tested on mRNA sequences from LncATLAS to assess whether lncRNA‑trained features generalize to protein‑coding transcripts. Performance varied by cell line: Cell Line AUROC (lncRNA) AUROC (mRNA) A549 0.761 0.800 H1.hESC 0.708 0.642 HeLa.S3 0.779 0.648 HUVEC 0.757 0.598 MCF.7 0.724 0.653 The model shows moderate cross‑predictivity but performs best for lncRNAs, as intended. For mRNA‑specific localization prediction, dedicated methods are recommended. Data Curation & Quality Control: Source: LncATLAS database (RNA‑seq, CNRCI values) + ENSEMBL (sequences, version 112) Length filter: ≤ 10,000 nucleotides (longer lncRNAs removed to reduce variability) Redundancy reduction: MeshClust at 90% sequence identity Labeling: CNRCI > 0 → cytoplasm; CNRCI < 0 → nucleus; CNRCI = 0 excluded Train/validation split: Stratified by class (80/20) Quality exclusion: Removed lncRNAs with no sequence information Usage: These datasets and models are designed for: Training cell‑line‑specific lncRNA subcellular localization predictors Understanding sequence determinants (correlation‑based features) of cytoplasm vs. nuclear retention Developing RNA‑based therapeutics (ASOs for nuclear targets; siRNAs for cytoplasmic targets) Guiding CRISPR strategies (CRISPRi/Cas9 for nuclear lncRNAs; Cas13 for cytoplasmic lncRNAs) Functional annotation of newly discovered lncRNAs Therapeutic Rationale (from paper): Nuclear lncRNAs → Antisense oligonucleotides (ASOs) or CRISPRi/Cas9 (disrupt transcription) Cytoplasmic lncRNAs → siRNAs or CRISPR‑Cas13 (degrade transcript directly) Related Resources: Web server: https://webs.iiitd.edu.in/raghava/cytolncpred/ GitHub: https://github.com/raghavagps/cytolncpred Nfeature (feature computation tool): https://webs.iiitd.edu.in/raghava/nfeature/ License: CC BY 4.0 Contact: Prof. Gajendra P. S. Raghava Files raghavagps/cytolncpred-v1.0.zip Files (17.5 MB) Name Size Download all raghavagps/cytolncpred-v1.0.zip md5:2382ffa2b1dd8db04dcf4ebba13da2a5 17.5 MB Preview Download Additional details Related works Is supplement to Software: https://github.com/raghavagps/cytolncpred/tree/v1.0 (URL) Software Repository URL https://github.com/raghavagps/cytolncpred 16 Views 7 Downloads Show more details All versions This version Views Total views 16 16 Downloads Total downloads 7 7 Data volume Total data volume 122.3 MB 122.3 MB More info on how stats are collected.... Versions External resources Archived in Software Heritage swh:1:dir:bd60c37f8350eaccb6c585b17f4a706550e0236a Available in raghavagps/cytolncpred Release: v1.0 Indexed in OpenAIRE Communities Details DOI DOI Badge DOI 10.5281/zenodo.19877518 Markdown [](https://doi.org/10.5281/zenodo.19877518) reStructuredText .. image:: https://zenodo.org/badge/DOI/10.5281/zenodo.19877518.svg :target: https://doi.org/10.5281/zenodo.19877518 HTML <a href="https://doi.org/10.5281/zenodo.19877518"><img src="https://zenodo.org/badge/DOI/10.5281/zenodo.19877518.svg" alt="DOI"></a> Image URL https://zenodo.org/badge/DOI/10.5281/zenodo.19877518.svg Target URL https://doi.org/10.5281/zenodo.19877518 Resource type Software Publisher Zenodo Rights License GNU General Public License v3.0 only No further description. Read more Citation Export Technical metadata Created April 29, 2026 Modified April 29, 2026 Jump up About About Policies Infrastructure Principles Projects Roadmap Contact Blog Blog Support Help FAQ Developers REST API OAI-PMH Contribute GitHub Donate Funded by Powered by CERN Data Centre & InvenioRDM Status Privacy policy Cookie policy Terms of Use This site uses cookies. Find out more on how we use cookies Accept all cookies Accept only essential cookies

Related documents

• ANLN: A New Hub in Glutamine Metabolism of Lung Adenocarcinoma by scRNA-Seq and Machine Learning.#402577
• UNCOVERING LNCRNA-CHROMATIN INTERACTIONS SHAPING HUMAN MONOCYTE TO MACROPHAGE DIFFERENTIATION#274683
• Utilising Machine Learning and Single-Cell Analysis to Uncover SKCM Metastasis-Related Genes.#3732
• Machine learning approach to single cell transcriptomic analysis of Sjogren's disease reveals altered activation states of B and T lymphocytes.#247646
• Response to "Addressing biases and limitations in feature attribution for circRNA modification profiling".#3928