D-Flow: Multi-modality Flow Matching for D-peptide Design

• DOI: 10.48550/arxiv.2411.10618
• Publication Date: 2024-11-15
• Authors: Fang Wu; Tianchao Xu; S. Jin; Xiangru Tang; N. Xu; James Zou; Brian Hie
• Keywords: Modality (human–computer interaction)

Proteins play crucial roles in biological processes, with therapeutic peptides emerging as promising pharmaceutical agents. They allow new possibilities to leverage target binding sites that were previously undruggable. While deep learning (DL) has advanced peptide discovery, generating D-proteins composed of D-amino acids remains challenging due the scarcity natural examples. This paper proposes D-Flow, a full-atom flow-based framework for {de novo} D-peptide design. D-Flow is conditioned on receptor and utilizes comprehensive representation structure, incorporating backbone frames, side-chain angles, discrete amino acid types. A mirror-image algorithm implemented address lack training data D-proteins, which converts L-receptors' chirality. Furthermore, we enhance D-Flow's capacity by integrating large protein language models (PLMs) structural awareness through lightweight adapter. two-stage pipeline controlling toolkit also enable transition from general design targeted binder while preserving pretraining knowledge. Extensive experimental results PepMerge benchmark demonstrate effectiveness, particularly developing entire D-residues. approach represents significant advancement computational design, offering unique opportunities bioorthogonal stable molecular tools diagnostics. The code available in~\url{https://github.com/smiles724/PeptideDesign}.