[2L1-OS-25] OS-25

[2L1-OS-25-03] PLM-Guided Reinforcement Learning for Functional Protein Design

〇Shota Yamamoto¹, Jianming Huang¹, Hiroyuki Kasai¹ (1. Waseda University)

Keywords:Protein Language Model, Protein Design, Reinforcement Learning

In the design of functional proteins, conventional methods employing reinforcement learning and function prediction models often overlook the biological plausibility of generated sequences. Recent approaches that leverage protein language models (PLMs) have achieved great success in biological experiments. However, the supervised fine-tuning is required to enhance the functionality of PLMs, which is difficult because of its significant computational costs. In this study, we introduce a novel method of a PLM-guided reinforcement learning model, aiming to achieve better functionality and biological plausibility without fine-tuning. Experiments on real-world datasets demonstrate that proteins designed by our proposed approach can achieve better functionality.