Conditional Protein Structure Generation with Protpardelle-1c

Abstract We present Protpardelle-1c, a collection of protein structure generative models with robust motif scaffolding and support for multi-chain complex generation under hotspot-conditioning. Enabling sidechain-conditioning to a backbone-only model increased Protpardelle-1c’s MotifBench score from 4.97 to 28.16, outperforming RFdiffusion’s 21.27. The crop-conditional all-atom model achieved 208 unique solutions on the La-Proteina all-atom motif scaffolding benchmark, on par with La-Proteina while having ∼10 times fewer parameters. At 22M parameters, Protpardelle-1c enables rapid sampling, taking 40 minutes to sample all 3000 MotifBench backbones on an NVIDIA A100-80GB, compared to 31 hours for RFdiffusion. Competing Interest Statement The authors have declared no competing interest. Footnotes ↵† Work done as visiting researcher at Stanford University tianyulu{at}stanford.edu rshuai{at}stanford.edu petrkouba3{at}gmail.com zhaoyangli{at}stanford.edu yilinc5{at}stanford.edu akios{at}stanford.edu zhkim216{at}stanford.edu Author affiliation of Richard Shuai changed to "Biophysics Program" from "Department of Biophysics" Typo in Figure 4: La-Proteina (unindexed) changed to 357 from 354. Typo correction "is a based model" to "is a base model" Add "replacement guidance" to initial-release Protpardelle MotifBench hyperparameters. SHAPES citation updated to published Cell Systems version.