PPFlow: Target-aware Peptide Design with Torsional Flow Matching

TL;DR

PPFlow is a novel AI method for peptide drug design that models torsion angles using flow matching, enabling effective peptide generation and optimization, and is supported by a new protein-peptide dataset.

Contribution

The paper introduces PPFlow, a target-aware peptide design approach based on conditional flow matching on torus manifolds, and provides a new dataset PPBench2024 for structure-based peptide drug discovery.

Findings

PPFlow achieves state-of-the-art results in peptide drug generation.

PPFlow generalizes well to docking and side-chain packing tasks.

The PPBench2024 dataset supports training deep learning models for peptide design.

Abstract

Therapeutic peptides have proven to have great pharmaceutical value and potential in recent decades. However, methods of AI-assisted peptide drug discovery are not fully explored. To fill the gap, we propose a target-aware peptide design method called \textsc{PPFlow}, based on conditional flow matching on torus manifolds, to model the internal geometries of torsion angles for the peptide structure design. Besides, we establish a protein-peptide binding dataset named PPBench2024 to fill the void of massive data for the task of structure-based peptide drug design and to allow the training of deep learning methods. Extensive experiments show that PPFlow reaches state-of-the-art performance in tasks of peptide drug generation and optimization in comparison with baseline models, and can be generalized to other tasks including docking and side-chain packing.