PEPEVOLVE: POSITION-AWARE DYNAMIC PEPTIDE OPTIMIZATION VIA GROUP-RELATIVE ADVANTAGE
1. A novel study introduces PepEVOLVE, a novel framework for optimizing macrocyclic peptides that dynamically identifies optimal editing sites and improves peptide properties through multi-objective optimization.
2. Unlike previous methods like PepINVENT that require pre-specifying mutable positions, PepEVOLVE uses a context-free multi-armed bandit router to automatically discover high-impact residues for modification, enhancing the efficiency and flexibility of peptide optimization.
3. The framework incorporates dynamic pretraining with strategies like dynamic masking and CHUCKLES shifting, which significantly improve the model's generalization ability and robustness to representational shifts in peptide sequences.
4. During optimization, PepEVOLVE employs an evolving algorithm that allows peptides to iteratively evolve, coupled with a group-relative advantage mechanism to stabilize reinforcement learning updates and explore diverse peptide design spaces.
5. In silico evaluations demonstrate PepEVOLVE's superior performance, achieving higher mean scores and identifying better candidates compared to PepINVENT, while converging faster under tasks like optimizing permeability and lipophilicity with structural constraints.
6. This work offers a practical and reproducible solution for peptide lead optimization when optimal edit sites are unknown, enabling more efficient exploration and improving design quality across multiple objectives, thus advancing computational peptide design.
📜Paper:
arxiv.org/abs/2511.16912v1
#PepEVOLVE #PeptideOptimization #ComputationalBiology #DynamicOptimization #MachineLearning #DrugDiscovery