{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2026,10,10]],"date-time":"2026-10-10T01:51:33Z","timestamp":1791597093873,"version":"4.3.4"},"reference-count":61,"publisher":"Oxford University Press (OUP)","issue":"5","license":[{"start":{"date-parts":[[2026,10,10]],"date-time":"2026-10-10T00:00:00Z","timestamp":1791590400000},"content-version":"vor","delay-in-days":39,"URL":"https:\/\/creativecommons.org\/licenses\/by\/4.0\/"}],"funder":[{"DOI":"10.13039\/501100001809","name":"National Natural Science Foundation of China","doi-asserted-by":"publisher","award":["62306290"],"award-info":[{"award-number":["62306290"]}],"id":[{"id":"10.13039\/501100001809","id-type":"DOI","asserted-by":"publisher"}]}],"content-domain":{"domain":[],"crossmark-restriction":false},"short-container-title":[],"published-print":{"date-parts":[[2026,9,1]]},"abstract":"<jats:title>Abstract<\/jats:title>\n                  <jats:p>Multimodal protein language models (PLMs) deliver strong performance on mutation-effect prediction, but training such models from scratch demands substantial computational resources. In this paper, we propose a fine-tuning framework called InstructPLM-mu and investigate the question: Can a pretrained sequence-only PLM, augmented with structural information during fine-tuning, match the performance of end-to-end trained multimodal models? Surprisingly, our experiments show that fine-tuning ESM2 with structural inputs can reach performance comparable to ESM3. To understand how this is achieved, we systematically compare three different feature-fusion designs and fine-tuning recipes. Our results reveal that both the fusion method and the tuning strategy strongly affect final accuracy, indicating that the fine-tuning process is not trivial. We hope this work offers practical guidance for injecting structure into pretrained PLMs and motivates further research on better fusion mechanisms and fine-tuning protocols.<\/jats:p>","DOI":"10.1093\/bib\/bbag400","type":"journal-article","created":{"date-parts":[[2026,10,7]],"date-time":"2026-10-07T11:52:38Z","timestamp":1791373958000},"source":"Crossref","is-referenced-by-count":0,"title":["InstructPLM-mu: protein fitness prediction from structure instructions"],"prefix":"10.1093","volume":"27","author":[{"given":"Junde","family":"Xu","sequence":"first","affiliation":[{"name":"Department of Computer Science and Engineering, The Chinese University of Hong Kong , Shatin, New Territory, Hong Kong SAR, 999077,","place":["China"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Yapin","family":"Shi","sequence":"additional","affiliation":[{"name":"Hangzhou Institute for Advanced Study, University of Chinese Academy of Sciences , No. 1, Xiangshan Zhinong, Xihu District, Hangzhou, Zhejiang, 310024,","place":["China"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Lijun","family":"Lang","sequence":"additional","affiliation":[{"name":"Department of Computer Science and Engineering, The Chinese University of Hong Kong , Shatin, New Territory, Hong Kong SAR, 999077,","place":["China"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Taoyong","family":"Cui","sequence":"additional","affiliation":[{"name":"Department of Computer Science and Engineering, The Chinese University of Hong Kong , Shatin, New Territory, Hong Kong SAR, 999077,","place":["China"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Zhiming","family":"Zhang","sequence":"additional","affiliation":[{"name":"Hangzhou Institute of Medicine, Chinese Academy of Sciences , 150 Dongfang Street Xiasha, Qiantang District, Hangzhou, Zhejiang, 310000,","place":["China"]},{"name":"College of Intelligence and Computing, Tianjin University , No. 135 Yaguan Road, Haihe Education Park, Tianjin, 300350,","place":["China"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Guangyong","family":"Chen","sequence":"additional","affiliation":[{"name":"Hangzhou Institute of Medicine, Chinese Academy of Sciences , 150 Dongfang Street Xiasha, Qiantang District, Hangzhou, Zhejiang, 310000,","place":["China"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Jiezhong","family":"Qiu","sequence":"additional","affiliation":[{"name":"Hangzhou Institute of Medicine, Chinese Academy of Sciences , 150 Dongfang Street Xiasha, Qiantang District, Hangzhou, Zhejiang, 310000,","place":["China"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Pheng-Ann","family":"Heng","sequence":"additional","affiliation":[{"name":"Department of Computer Science and Engineering, The Chinese University of Hong Kong , Shatin, New Territory, Hong Kong SAR, 999077,","place":["China"]}],"role":[{"vocabulary":"crossref","role":"author"}]}],"member":"286","published-online":{"date-parts":[[2026,10,9]]},"reference":[{"key":"2026100921333158100_ref1","doi-asserted-by":"publisher","first-page":"1120370","DOI":"10.3389\/fbinf.2023.1120370","article-title":"Before and after AlphaFold2: an overview of protein structure prediction","volume":"3","author":"Bertoline","year":"2023","journal-title":"Front Bioinform"},{"key":"2026100921333158100_ref2","doi-asserted-by":"publisher","first-page":"620","DOI":"10.1038\/s41596-024-01060-5","article-title":"Easy and accurate protein structure prediction using colabfold","volume":"20","author":"Kim","year":"2025","journal-title":"Nat Protoc"},{"key":"2026100921333158100_ref3","doi-asserted-by":"publisher","first-page":"275","DOI":"10.1038\/s41587-023-01763-2","article-title":"Efficient evolution of human antibodies from general protein language models","volume":"42","author":"Hie","year":"2024","journal-title":"Nat Biotechnol"},{"key":"2026100921333158100_ref4","doi-asserted-by":"crossref","first-page":"W239","DOI":"10.1093\/nar\/gkl190","article-title":"CUPSAT: prediction of protein stability upon point mutations","volume":"34","author":"Vijaya Parthiban","year":"2006","journal-title":"Nucleic Acids Res"},{"key":"2026100921333158100_ref5","doi-asserted-by":"publisher","first-page":"116","DOI":"10.1038\/s42256-020-0149-6","article-title":"A topology-based network tree for the prediction of protein\u2013protein binding affinity changes following mutation","volume":"2","author":"Wang","year":"2020","journal-title":"Nat Mach Intell"},{"key":"2026100921333158100_ref6","doi-asserted-by":"publisher","first-page":"789","DOI":"10.1038\/nchembio.232","article-title":"The role of dynamic conformational ensembles in biomolecular recognition","volume":"5","author":"Boehr","year":"2009","journal-title":"Nat Chem Biol"},{"key":"2026100921333158100_ref7","doi-asserted-by":"publisher","first-page":"5543","DOI":"10.1021\/acs.jpclett.4c00017","article-title":"Dynamic personality of proteins and effect of the molecular environment","volume":"15","author":"Sonaglioni","year":"2024","journal-title":"J Phys Chem Lett"},{"key":"2026100921333158100_ref8","doi-asserted-by":"publisher","DOI":"10.1038\/s43586-025-00383-1","article-title":"Computational protein design","volume":"5","author":"Albanese","year":"2025","journal-title":"Nat Rev Methods Primers"},{"key":"2026100921333158100_ref9","doi-asserted-by":"publisher","first-page":"2267","DOI":"10.1038\/nprot.2014.153","article-title":"Measuring the activity of protein variants on a large scale using deep mutational scanning","volume":"9","author":"Fowler","year":"2014","journal-title":"Nat Protoc"},{"key":"2026100921333158100_ref10","doi-asserted-by":"publisher","first-page":"801","DOI":"10.1038\/nmeth.3027","article-title":"Deep mutational scanning: a new style of protein science","volume":"11","author":"Fowler","year":"2014","journal-title":"Nat Methods"},{"key":"2026100921333158100_ref11","doi-asserted-by":"publisher","first-page":"123","DOI":"10.1016\/j.tips.2021.11.010","article-title":"Deep mutational scanning for therapeutic antibody engineering","volume":"43","author":"Hanning","year":"2022","journal-title":"Trends Pharmacol Sci"},{"key":"2026100921333158100_ref12","doi-asserted-by":"publisher","first-page":"1123","DOI":"10.1126\/science.ade2574","article-title":"Evolutionary-scale prediction of atomic-level protein structure with a language model","volume":"379","author":"Lin","year":"2023","journal-title":"Science"},{"key":"2026100921333158100_ref13","doi-asserted-by":"publisher","DOI":"10.1093\/bioinformatics\/btae621","article-title":"Expert-guided protein language models enable accurate and blazingly fast fitness prediction","volume":"40","author":"Marquet","year":"2024","journal-title":"Bioinformatics"},{"key":"2026100921333158100_ref14","volume-title":"Proceedings of the Twelfth International Conference on Learning Representations (ICLR 2024); 2024 May 7\u201311; Vienna, Austria","author":""},{"key":"2026100921333158100_ref15","doi-asserted-by":"crossref","first-page":"101456","DOI":"10.52202\/079017-3217","article-title":"Multi-scale representation learning for protein fitness prediction","volume":"37","author":"Zhang","year":"2024","journal-title":"Adv Neural Inf Proces Syst"},{"key":"2026100921333158100_ref16","doi-asserted-by":"publisher","first-page":"100750","DOI":"10.1016\/j.xinn.2024.100750","article-title":"Structure-based self-supervised learning enables ultrafast protein stability prediction upon mutation","volume":"6","author":"Sun","year":"2025","journal-title":"Int J Hydrog Energ"},{"key":"2026100921333158100_ref17","first-page":"29287","article-title":"Language models enable zero-shot prediction of the effects of mutations on protein function","volume":"34","author":"Meier","year":"2021","journal-title":"Adv Neural Inf Proces Syst"},{"key":"2026100921333158100_ref18","author":"","journal-title":"Proceedings of the 42nd International Conference on Machine Learning; 2025 Jul 13\u201319; Vancouver, Canada"},{"key":"2026100921333158100_ref19","doi-asserted-by":"crossref","first-page":"i401","DOI":"10.1093\/bioinformatics\/btaf189","article-title":"From high-throughput evaluation to wet-lab studies: advancing mutation effect prediction with a retrieval-enhanced model","volume":"41","author":"Tan","year":"2025","journal-title":"Bioinformatics"},{"key":"2026100921333158100_ref20","volume-title":"NeurIPS 2024 Workshop on AI for New Drug Modalities; 2024 Dec 15; Vancouver, Canada","author":""},{"key":"2026100921333158100_ref21","doi-asserted-by":"crossref","first-page":"64331","DOI":"10.52202\/075280-2810","article-title":"ProteinGym: large-scale benchmarks for protein fitness prediction and design","volume":"36","author":"Notin","year":"2023","journal-title":"Adv Neural Inf Proces Syst"},{"key":"2026100921333158100_ref22","doi-asserted-by":"publisher","volume-title":"Nat Biotechnol","author":"","DOI":"10.1038\/s41587-025-02859-7"},{"key":"2026100921333158100_ref23","volume-title":"Proceedings of the 40th International Conference on Machine Learning; 2023 Jul 23\u201329; Honolulu, Hawaii, USA","author":""},{"key":"2026100921333158100_ref24","doi-asserted-by":"publisher","author":"","DOI":"10.1101\/2024.04.17.589642"},{"key":"2026100921333158100_ref25","doi-asserted-by":"publisher","author":"","DOI":"10.1101\/2024.08.03.606485"},{"key":"2026100921333158100_ref26","doi-asserted-by":"publisher","first-page":"49","DOI":"10.1126\/science.add2187","article-title":"Robust deep learning\u2013based protein sequence design using ProteinMPNN","volume":"378","author":"Dauparas","year":"2022","journal-title":"Science"},{"key":"2026100921333158100_ref27","volume-title":"Proceedings of the 39th International Conference on Machine Learning; 2022 Jul 17\u201323; Baltimore, Maryland, USA","author":""},{"key":"2026100921333158100_ref28","doi-asserted-by":"publisher","first-page":"850","DOI":"10.1126\/science.ads0018","article-title":"Simulating 500 million years of evolution with a language model","volume":"387","author":"Hayes","year":"2025","journal-title":"Science"},{"key":"2026100921333158100_ref29","doi-asserted-by":"publisher","DOI":"10.1038\/s41467-022-30936-x","article-title":"The 3D mutational constraint on amino acid sites in the human proteome","volume":"13","author":"Li","year":"2022","journal-title":"Nat Commun"},{"key":"2026100921333158100_ref30","doi-asserted-by":"publisher","DOI":"10.1126\/science.adg7492","article-title":"Accurate proteome-wide missense variant effect prediction with AlphaMissense","volume":"381","author":"Cheng","year":"2023","journal-title":"Science"},{"key":"2026100921333158100_ref31","doi-asserted-by":"publisher","first-page":"3031","DOI":"10.1021\/acs.jctc.7b00125","article-title":"The Rosetta all-atom energy function for macromolecular modeling and design","volume":"13","author":"Alford","year":"2017","journal-title":"J Chem Theory Comput"},{"key":"2026100921333158100_ref32","doi-asserted-by":"publisher","first-page":"136","DOI":"10.1093\/bioinformatics\/btz515","article-title":"Increasing the efficiency and accuracy of the abacus protein sequence design method","volume":"36","author":"Xiong","year":"2020","journal-title":"Bioinformatics"},{"key":"2026100921333158100_ref33","doi-asserted-by":"crossref","first-page":"35700","DOI":"10.52202\/079017-1126","article-title":"ProSST: protein language modeling with quantized structure and disentangled attention","volume":"37","author":"Li","year":"2024","journal-title":"Adv Neural Inf Proces Syst"},{"key":"2026100921333158100_ref34","doi-asserted-by":"publisher","DOI":"10.1093\/bioinformatics\/btaf189","volume":"41","author":"","journal-title":"Bioinformatics"},{"key":"2026100921333158100_ref35","doi-asserted-by":"crossref","first-page":"423","DOI":"10.1109\/TPAMI.2018.2798607","article-title":"Multimodal machine learning: a survey and taxonomy","volume":"41","author":"Baltru\u0161aitis","year":"2018","journal-title":"IEEE Trans Pattern Anal Mach Intell"},{"key":"2026100921333158100_ref36","doi-asserted-by":"crossref","first-page":"23716","DOI":"10.52202\/068431-1723","article-title":"Flamingo: a visual language model for few-shot learning","volume":"35","author":"Alayrac","year":"2022","journal-title":"Adv Neural Inf Proces Syst"},{"key":"2026100921333158100_ref37","doi-asserted-by":"crossref","first-page":"87310","DOI":"10.52202\/079017-2771","article-title":"Cambrian-1: a fully open, vision-centric exploration of multimodal LLMs","volume":"37","author":"Tong","year":"2024","journal-title":"Adv Neural Inf Proces Syst"},{"key":"2026100921333158100_ref38","doi-asserted-by":"publisher","DOI":"10.1007\/s11263-025-02667-1","volume":"134","author":"","journal-title":"Int J Comput Vis"},{"key":"2026100921333158100_ref39","doi-asserted-by":"crossref","first-page":"34892","DOI":"10.52202\/075280-1516","article-title":"Visual instruction tuning","volume":"36","author":"Liu","year":"2023","journal-title":"Adv Neural Inf Proces Syst"},{"key":"2026100921333158100_ref40","doi-asserted-by":"publisher","volume-title":"Proceedings of the IEEE\/CVF Conference on Computer Vision and Pattern Recognition; 2024 Jun 16\u201322; Seattle, WA, USA","author":"","DOI":"10.1109\/CVPR52733.2024.02484"},{"key":"2026100921333158100_ref41","author":""},{"key":"2026100921333158100_ref42","author":""},{"key":"2026100921333158100_ref43","author":""},{"key":"2026100921333158100_ref44","doi-asserted-by":"crossref","first-page":"23464","DOI":"10.52202\/079017-0739","article-title":"DeepStack: deeply stacking visual tokens is surprisingly simple and effective for LMMs","volume":"37","author":"Meng","year":"2024","journal-title":"Adv Neural Inf Proces Syst"},{"key":"2026100921333158100_ref45","doi-asserted-by":"publisher","author":"","journal-title":"Proceedings of the IEEE\/CVF Conference on Computer Vision and Pattern Recognition; 2025 Jun 11\u201315; Nashville, TN, USA","DOI":"10.1109\/CVPR52734.2025.00393"},{"key":"2026100921333158100_ref46","author":""},{"key":"2026100921333158100_ref47"},{"key":"2026100921333158100_ref48","author":""},{"key":"2026100921333158100_ref49","doi-asserted-by":"publisher","author":"","journal-title":"Findings of the Association for Computational Linguistics: ACL 2024; 2024 Aug 11\u201316; Bangkok, Thailand","DOI":"10.18653\/v1\/2024.findings-acl.775"},{"key":"2026100921333158100_ref50","volume":"267","author":"","journal-title":"Proceedings of the 42nd International Conference on Machine Learning; 2025 Jul 13\u201319; Vancouver, Canada"},{"key":"2026100921333158100_ref51","doi-asserted-by":"publisher","first-page":"D266","DOI":"10.1093\/nar\/gkaa1079","article-title":"CATH: increased structural coverage of functional space","volume":"49","author":"Sillitoe","year":"2021","journal-title":"Nucleic Acids Res"},{"key":"2026100921333158100_ref52","doi-asserted-by":"publisher","DOI":"10.1093\/bioadv\/vbaf117","volume":"5","author":"","journal-title":"Bioinform Adv"},{"key":"2026100921333158100_ref53","doi-asserted-by":"publisher","first-page":"gzad015","DOI":"10.1093\/protein\/gzad015","article-title":"Masked inverse folding with sequence transfer for protein representation learning","volume":"36","author":"Yang","year":"2023","journal-title":"Protein Eng Des Sel"},{"key":"2026100921333158100_ref54","doi-asserted-by":"crossref","first-page":"RP98033","DOI":"10.7554\/eLife.98033.4","article-title":"Semantical and geometrical protein encoding toward enhanced bioactivity and thermostability","volume":"13","author":"Tan","year":"2025","journal-title":"Elife"},{"key":"2026100921333158100_ref55","first-page":"3","article-title":"LoRA: low-rank adaptation of large language models","volume":"1","author":"Hu","year":"2022","journal-title":"ICLR"},{"key":"2026100921333158100_ref56","author":""},{"key":"2026100921333158100_ref57","doi-asserted-by":"publisher","author":"","journal-title":"Proceedings of the IEEE\/CVF International Conference on Computer Vision; 2025 Oct 19\u201323; Honolulu, Hawaii, USA","DOI":"10.1109\/ICCV51701.2025.00009"},{"key":"2026100921333158100_ref58","doi-asserted-by":"publisher","volume-title":"Computer Vision \u2013 ECCV 2024: 18th European Conference, Proceedings, Part XVI; 2024 Sep 29\u2013Oct 4; Milan, Italy","author":"","DOI":"10.1007\/978-3-031-72640-8_7"},{"key":"2026100921333158100_ref59","doi-asserted-by":"publisher","volume-title":"Computer Vision \u2013 ECCV 2024: 18th European Conference, Proceedings, Part LXII; 2024 Sep 29\u2013Oct 4; Milan, Italy","author":"","DOI":"10.1007\/978-3-031-73033-7_3"},{"key":"2026100921333158100_ref60","author":"","journal-title":"The Thirteenth International Conference on Learning Representations (ICLR 2025); 2025 Apr 24\u201328; Singapore"},{"key":"2026100921333158100_ref61","author":""}],"container-title":["Briefings in Bioinformatics"],"original-title":[],"language":"en","link":[{"URL":"https:\/\/academic.oup.com\/bib\/article-pdf\/27\/5\/bbag400\/71793021\/bbag400.pdf","content-type":"application\/pdf","content-version":"vor","intended-application":"syndication"},{"URL":"https:\/\/academic.oup.com\/bib\/article-pdf\/27\/5\/bbag400\/71793021\/bbag400.pdf","content-type":"unspecified","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2026,10,10]],"date-time":"2026-10-10T01:33:42Z","timestamp":1791596022000},"score":1,"resource":{"primary":{"URL":"https:\/\/academic.oup.com\/bib\/article\/doi\/10.1093\/bib\/bbag400\/8899282"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2026,9,1]]},"references-count":61,"journal-issue":{"issue":"5","published-print":{"date-parts":[[2026,9,1]]}},"URL":"https:\/\/doi.org\/10.1093\/bib\/bbag400","relation":{},"ISSN":["1467-5463","1477-4054"],"issn-type":[{"value":"1467-5463","type":"print"},{"value":"1477-4054","type":"electronic"}],"subject":[],"published-other":{"date-parts":[[2026,9]]},"published":{"date-parts":[[2026,9,1]]},"article-number":"bbag400"}}