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

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: $\mathit{\text{Can}}\mathit{\text{multimodal}}\mathit{\text{fine-tuning}}\mathit{\text{of}}\mathit{\text{a}}\mathit{\text{pretrained,}}\mathit{\text{sequence-only}}\mathit{\text{protein}}\mathit{\text{language}}\mathit{\text{model}}\mathit{\text{match}}\mathit{\text{the}}\mathit{\text{performance}}\mathit{\text{of}}\mathit{\text{models}}\mathit{\text{trained}}\mathit{\text{end-to-end?}}$ 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.