Addressing flaws in the Seq2Topt dataset for the prediction of enzyme optimal temperature.

Addressing flaws in the Seq2Topt dataset for the prediction of enzyme optimal temperature - PMC Skip to main content An official website of the United States government Here's how you know Here's how you know Search Log in Primary site navigation Search Log in Search PMC Full-Text Archive Search in PMC Journal List User Guide 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 letter Brief Bioinform . 2026 May 31;27(3):bbag282. doi: 10.1093/bib/bbag282 Addressing flaws in the Seq2Topt dataset for the prediction of enzyme optimal temperature Peng Zhou 1, 2, ✉ Author information Article notes Copyright and License information 1 State Key Laboratory of Submarine Geoscience, Second Institute of Oceanography, Ministry of Natural Resources, No. 36 North Bao Chu Road, Xihu District, 310012 Hangzhou, China 2 Key Laboratory of Marine Ecosystem Dynamics, Ministry of Natural Resources and Second Institute of Oceanography, Ministry of Natural Resources, No. 36 North Bao Chu Road, Xihu District, 310012 Hangzhou, China ✉ Corresponding author. Second Institute of Oceanography, Ministry of Natural Resources, No. 36 North Bao Chu Road, Xihu District, 310012 Hangzhou, China. E-mail: zhoupeng@sio.org.cn Received 2026 May 3; Accepted 2026 May 11; Collection date 2026 May. © The Author(s) 2026. Published by Oxford University Press. This is an Open Access article distributed under the terms of the Creative Commons Attribution License ( https://creativecommons.org/licenses/by/4.0/ ), which permits unrestricted reuse, distribution, and reproduction in any medium, provided the original work is properly cited. PMC Copyright notice PMCID: PMC13222526  PMID: 42218709 To the Editor, I read the article from Qiu et al. with great interest, in which the authors developed a deep learning predictor, Seq2Topt, to estimate the optimal temperature ( T opt ) of enzymes. Seq2Topt achieved a prediction performance with an RMSE of 12.26°C using only protein sequence information [ 1 ]. Temperature is one of the critical factors governing enzyme catalytic activity [ 2 ]. A comprehensive understanding of T opt facilitates enzyme modification and industrial utilization through accurate T opt prediction. Accordingly, the development of Seq2Topt provides a promising computational tool for enzyme mining and in silico enzyme design. Nevertheless, although Seq2Topt yielded an RMSE of 12.26°C, the predictive performance still has considerable room for improvement. Improved model accuracy would enhance its practical value and applicability in real industrial and research scenarios. As the overall performance of deep learning models is fundamentally determined by the quality of their underlying datasets, I conducted a careful data inspection and verification. The relevant raw files were downloaded from the corresponding GitHub repository of Seq2Topt ( https://github.com/SizheQiu/Seq2Topt ). The training set accounted for 90% of the total dataset. I sorted the data in the file named “train.csv” in ascending order of annotated T opt values and selected nine entries with low T opt values (<15°C) for literature cross-validation and manual inspection. Surprisingly, rigorous data verification revealed that all nine selected entries were problematic. These nine anomalous records are summarized in Table 1 , and the underlying causes of the inconsistent or missing experimental evidence are discussed in detail below. Table 1. Nine anomalous entries found from Seq2Topt Dataset. UniProt accession number T opt in Seq2Topt dataset (°C) T opt from literature (°C) F9FU71 0 Not found A9U908 4 Not found J7HET3 4 Not found P94368 5 Not found Q5YXQ1 7.5 Not found Q83V33 10 Not found Q50539 13 Not found Q50538 13 Not found Q2WCS9 13 37 Open in a new tab F9FU71 is a linoleate diol synthase from Fusarium oxysporum . Its T opt was annotated as 0°C in the Seq2Topt training set [ 1 ]. Nevertheless, available experimental study only indicated incubation on ice during enzyme assay [ 3 ], with no credible evidence supporting a T opt of 0°C. A9U908 and J7HET3 encode superoxide dismutase sodA and sodB in Yersinia enterocolitica , respectively. Both enzymes were assigned a T opt of 4°C in the Seq2Topt training set. Although high enzymatic activity at 4°C had been documented in the literature [ 4 ], no activity measurements at lower temperatures were available. Accordingly, the reliability of the annotated T opt of 4°C remains questionable. P94368 is an ADP-dependent NAD(P)H-hydrate dehydratase from Bacillus subtilis , with a labeled T opt of 5°C in the Seq2Topt training set. After comprehensive literature retrieval [ 5–7 ], no experimental evidence or conclusive statement was found to confirm a T opt of 5°C for this enzyme. Q5YXQ1 is a maleate isomerase from Nocardia farcinica , recorded with a T opt of 7.5°C in the Seq2Topt training set. However, no study has officially defined its T opt as 7.5°C; only an optimal pH of 7.5 for this enzyme was reported [ 8 ]. Hence, the authenticity of this T opt annotation requires further verification. Q83V33 is an aminomuconate-semialdehyde dehydrogenase from Pseudomonas fluorescens . Seq2Topt annotated its T opt as 10°C. In the relevant study, enzyme activity assays were merely conducted at 10°C [ 9 ], whereas this temperature was not determined as the optimum. No experimentally validated T opt was found. Q50538 and Q50539 represent two subunits of the corrinoid/iron–sulfur enzyme of the CO dehydrogenase complex from Methanosarcina thermophila , with a unified annotated T opt of 13°C in the Seq2Topt training set. Similar to Q83V33 , the relevant literature only confirmed that multiple protein purification steps were carried out at 13°C for heterologously expressed proteins [ 10 ], rather than defining 13°C as their T opt . Q2WCS9 is a glutamate dehydrogenase (GDH) from Lactiplantibacillus plantarum , annotated with a T opt of 13°C in the Seq2Topt training set. In contrast, a relevant study showed that the GDH activity of the cytoplasm fraction of Lactiplantibacillus plantarum strains tested reached its optimum at 37°C [ 11 ], revealing a remarkable data discrepancy. In summary, all nine entries inspected were identified as anomalous through literature cross-checking. The Seq2Topt study stated that its dataset ( n = 2917) was retrieved from the GitHub repository of TOMER [ 12 ], which was originally obtained from the BRENDA database [ 13 ]. Data quality fundamentally affects the training, optimization, and performance evaluation of machine learning models. To mitigate the above data-related issues, the existing T opt annotations should be thoroughly inspected and corrected. Upon careful verification of the T opt annotations, a more reliable dataset is expected. I hope these remarks contribute to further refinement of enzyme T opt prediction models and promote their more reliable practical applications. Conflicts of interest The author declares that there is no conflict of interest. Funding None declared. Data availability No new data were generated in this study. The letter refers to the data provided by Seq2Topt study [ 1 ]. References 1. Qiu  S, Hu  B, Zhao  J  et al. Seq2Topt: a sequence-based deep learning predictor of enzyme optimal temperature. Brief Bioinform 2025;26:bbaf114. 10.1093/bib/bbaf114 [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 2. Nelson  DL, Cox  MM, Hoskins  AA. Lehninger Principles of Biochemistry 8th edn. Austin: Macmillan Learning, 2021. [ Google Scholar ] 3. Sooman  L, Oliw  EH. Discovery of a novel linoleate dioxygenase of Fusarium oxysporum and linoleate diol synthase of Colletotrichum graminicola . Lipids 2015;50:1243–52. 10.1007/s11745-015-4078-9 [ DOI ] [ PubMed ] [ Google Scholar ] 4. Dhar  MS, Gupta  V, Virdi  JS. Detection, distribution and characterization of novel superoxide dismutases from Yersinia enterocolitica Biovar 1A.  Skurnik M, editor. PLoS One 2013;8:e63919. 10.1371/journal.pone.0063919 [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 5. Shumilin  IA, Cymborowski  M, Chertihin  O  et al. Identification of unknown protein function using metabolite cocktail screening. Structure 2012;20:1715–25. 10.1016/j.str.2012.07.016 [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 6. Petrovova  M, Tkadlec  J, Dvoracek  L  et al. NAD(P)H-hydrate dehydratase—a metabolic repair enzyme and its role in Bacillus subtilis stress adaptation.  Appanna VD, editor. PLoS One 2014;9:e112590. 10.1371/journal.pone.0112590 [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 7. Zhang  R, Grembecka  J, Vinokour  E  et al. Structure of Bacillus subtilis YXKO—a member of the UPF0031 family and a putative kinase. J Struct Biol 2002;139:161–70. 10.1016/S1047-8477(02)00532-4 [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 8. Fisch  F, Fleites  CM, Delenne  M  et al. A covalent succinylcysteine-like intermediate in the enzyme-catalyzed transformation of maleate to fumarate by maleate isomerase. J Am Chem Soc 2010;132:11455–7. 10.1021/ja1053576 [ DOI ] [ PubMed ] [ Google Scholar ] 9. Huo  L, Davis  I, Liu  F  et al. Crystallographic and spectroscopic snapshots reveal a dehydrogenase in action. Nat Commun 2015;6:5935. 10.1038/ncomms6935 [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 10. Maupin-Furlow  J, Ferry  JG. Characterization of the cdhD and cdhE genes encoding subunits of the corrinoid/iron-sulfur enzyme of the CO dehydrogenase complex from Methanosarcina thermophila . J Bacteriol 1996;178:340–6. 10.1128/jb.178.2.340-346.1996 [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 11. De Angelis  M, Calasso  M, Di Cagno  R  et al. NADP-glutamate dehydrogenase activity in nonstarter lactic acid bacteria: effects of temperature, pH and NaCl on enzyme activity and expression: GDH activity in nonstarter lactic acid bacteria. J Appl Microbiol 2010;109:1763–74. 10.1111/j.1365-2672.2010.04804.x [ DOI ] [ PubMed ] [ Google Scholar ] 12. Gado  JE, Beckham  GT, Payne  CM. Improving enzyme optimum temperature prediction with resampling strategies and ensemble learning. J Chem Inf Model 2020;60:4098–107. 10.1021/acs.jcim.0c00489 [ DOI ] [ PubMed ] [ Google Scholar ] 13. Schomburg  I, Jeske  L, Ulbrich  M  et al. The BRENDA enzyme information system—from a database to an expert system. J Biotechnol 2017;261:194–206. 10.1016/j.jbiotec.2017.04.020 [ DOI ] [ PubMed ] [ Google Scholar ] Associated Data This section collects any data citations, data availability statements, or supplementary materials included in this article. Data Availability Statement No new data were generated in this study. The letter refers to the data provided by Seq2Topt study [ 1 ]. Articles from Briefings in Bioinformatics are provided here courtesy of Oxford University Press ACTIONS View on publisher site PDF (350.6 KB) Cite Collections Permalink RESOURCES Similar articles Cited by other articles Links to NCBI Databases 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

Related documents

• EZPro-Multi: Contrastive Learning-Enhanced Multi-property Prediction for Enzyme Engineering#229660
• EINN: An enzyme-informed neural network guided by an enzyme-constrained genome-scale metabolic model.#457443
• EnzymeHunter: Achieving fine-grained enzyme function prediction with a hierarchically aware contrastive learning framework.#371547
• gbharti/finance-alpaca#786943
• Overcoming systematic data biases enables accurate prediction of enzyme <i>k</i> _cat fold-changes for computational protein design#273238