arXiv 2510.033702025-10q-bio.QMcs.AIcs.CE

InstructPLM-mu：在1小时内微调ESM2在蛋白质突变预测中优于ESM3

InstructPLM-mu: 1-Hour Fine-Tuning of ESM2 Beats ESM3 in Protein Mutation Predictions

The Chinese University of Hong Kong（香港中文大学）；Hangzhou Institute of Medicine, CAS（杭州医学研究所，中国科学院）；Hangzhou Institute for Advanced Study, UCAS（杭州先进研究 institute，中国科学院大学）；University of Chinese Academy of Sciences（中国科学院大学）；Tianjin University（天津大学）

Junde Xu, Yapin Shi, Lijun Lang, Taoyong Cui, Zhiming Zhang, Guangyong Chen, Jiezhong Qiu, Pheng-Ann Heng

备注：: preprint

阅读 PDF

arXiv 页面 ↗HTML 原文 ↗

AI总结：

InstructPLM-mu通过1小时微调ESM2在蛋白质突变预测中超越ESM3，揭示了结构输入对模型性能的关键影响。

摘要：

多模态蛋白质语言模型在突变效应预测上表现强劲，但从头训练此类模型需要大量计算资源。本文提出了一种名为InstructPLM-mu的微调框架，并试图回答一个问题：\textit{能否通过微调一个预训练的仅序列蛋白质语言模型来达到端到端训练模型的性能？}出人意料的是，我们的实验显示，使用结构输入微调ESM2可以达到与ESM3相当的性能。为了了解这一现象是如何实现的，我们系统地比较了三种不同的特征融合设计和微调方案。我们的结果表明，融合方法和微调策略对最终准确性有显著影响，这表明微调过程并不简单。我们希望这项工作能为将结构注入预训练的蛋白质语言模型提供实用指导，并激励进一步研究更好的融合机制和微调协议。

英文摘要：

Multimodal protein language models 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 try to answer a question: \textit{Can multimodal fine-tuning of a pretrained, sequence-only protein language model match the performance of models trained end-to-end? } 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 protein language models and motivates further research on better fusion mechanisms and fine-tuning protocols.

机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。