HighPlay: Cyclic Peptide Sequence Design Based on Reinforcement Learning and Protein Structure Prediction

The structural diversity and good biocompatibility of
cyclic peptides
have led to their emergence as potential therapeutic agents. Existing
cyclic peptide design methods, whether traditional or emerging AI-assisted,
rely on a multitude of experiments and face challenges such as limited
molecular diversity, high cost, and time-consuming. In this study,
we propose HighPlay, which integrates reinforcement learning (MCTS)
with the HighFold structural prediction model to design cyclic peptide
sequences based solely on the target protein sequence information,
to achieve the synergistic optimization of cyclic peptide sequences
and binding sites and to dynamically explore the sequence space without
the need for predefined target information. The model was applied
to the design of cyclic peptide sequences for three different targets,
which were screened and verified through molecular dynamics simulations,
demonstrating good binding affinity. Specifically, the cyclic peptide
sequences designed for the TEAD4 target exhibited micromolar-level
affinity in further experimental validation.

环肽（cyclic peptides）兼具结构多样性与优良生物相容性，使其成为极具潜力的治疗剂。现有环肽设计方法，无论是传统方案还是新兴的人工智能辅助技术，均依赖大量实验，且面临分子多样性受限、成本高昂、耗时冗长等诸多挑战。本研究提出HighPlay模型，该方法将强化学习（MCTS）与HighFold结构预测模型相结合，仅基于目标蛋白序列信息设计环肽序列，实现环肽序列与结合位点的协同优化，且无需预先设定靶点信息即可动态探索序列空间。该模型被应用于三类不同靶点的环肽序列设计，通过分子动力学模拟完成筛选与验证，结果显示其具备优良的结合亲和力。具体而言，针对TEAD4靶点设计的环肽序列在后续实验验证中展现出微摩尔级的结合亲和力。