Target-Specific <i>De Novo</i> Peptide Binder\nDesign with DiffPepBuilder

Despite the exciting progress in\ntarget-specific <i>de novo</i> protein binder design, peptide\nbinder design remains\nchallenging\ndue to the flexibility of peptide structures and the scarcity of protein-peptide\ncomplex structure data. In this study, we curated a large synthetic\ndata set, referred to as PepPC-F, from the abundant protein–protein\ninterface data and developed DiffPepBuilder, a <i>de novo</i> target-specific peptide binder generation method that utilizes an\nSE(3)-equivariant diffusion model trained on PepPC-F to codesign peptide\nsequences and structures. DiffPepBuilder also introduces disulfide\nbonds to stabilize the generated peptide structures. We tested DiffPepBuilder\non 30 experimentally verified strong peptide binders with available\nprotein–peptide complex structures. DiffPepBuilder was able\nto effectively recall the native structures and sequences of the peptide\nligands and to generate novel peptide binders with improved binding\nfree energy. We subsequently conducted <i>de novo</i> generation\ncase studies on three targets. In both the regeneration test and case\nstudies, DiffPepBuilder outperformed AfDesign and RFdiffusion coupled\nwith ProteinMPNN, in terms of sequence and structure recall, interface\nquality, and structural diversity. Molecular dynamics simulations\nconfirmed that the introduction of disulfide bonds enhanced the structural\nrigidity and binding performance of the generated peptides. As a general\npeptide binder <i>de novo</i> design tool, DiffPepBuilder\ncan be used to design peptide binders for given protein targets with\nthree-dimensional and binding site information.