ProteoCraft: Protein Design as Proposal-Verification Search

Abstract

Protein-protein interactions underpin nearly all biological processes. Yet, designing proteins with desired interactions remains challenging due to the structural complexity governing molecular recognition. This study views protein design as a search optimization problem over a combinatorial sequence space under expensive structural evaluation. Toward this goal, we introduce a novel tree-based search framework, ProteoCraft, that employs approximate structural gradients from diffusion-based structure predictors as effective proposal policies to specify where the search should explore. Complete no-gradient inference of folding models separately assigns values to the resulting sequences, determining what is good. ProteoCraft allocates this verification budget through a coarse-to-fine paradigm: multi-site edits first explore distinct sequence basins, then verified promising states undergo strict single-site refinement. Across fourteen de novo binder-design settings spanning eight target proteins, five immune-relevant TCR-facing targets, and four localized CDR redesign tasks, ProteoCraft improves design quality and high-confidence yield. These improvements persist under independent AF2-IG and Rosetta-based interface evaluation. Prospective experimental validation further yielded an 82.9% binding hit rate for PD-L1 miniproteins, with a best measured KD of 0.879 nM, and confirmed binding of designed IFNAR2 VHH. Our code is available at https://github.com/smiles724/ProteoCraft_v1.

In silico design of novel indolinone-based anticancer agents and proposal of synthetic routes.
Mohammed Er-Rajy, Mohamed El Fadili, Hanane Etabti +6
Frontiers in bioinformatics14 Oct 2026Peer-reviewedCC BYFull textMolecular biology & gene regulationEurope PMCPMC
Structural motif search across the protein universe with Folddisco
Hyunbin Kim, Rachel Seongeun Kim, Milot Mirdita +2
Nature Biotechnology5 Jun 2026Peer-reviewedLicence not statedFull text14 citedMolecular biology & gene regulationOpenAlex
Novel hybrid protein scaffold gap filling using weighted machine learning ensemble, beam search, and mass-constrained reranking
Tahmid Enam Shrestha, Md. Manzurul Hasan, Md. Rafiqul Islam
arXiv1 Sep 2026PreprintarXiv licenceFull textBiophysicsarXiv
SSAlign: fast and sensitive protein structure search at scale
Lei Wang, Xuchao Zhang, Yan Wang +1
Bioinformatics26 Oct 2026Peer-reviewedCC BY 4.0Full text1 citedMolecular biology & gene regulationPMC
De novo Rubisco design with protein language models
A. J. Kehl, S. K. S. Chu, J. H. Pereira +6
bioRxiv4 Sep 2026PreprintAll rights reservedFull textMolecular biology & gene regulationbioRxiv

Cite

BibTeX RIS

BibTeX

@misc{wu2026proteocraft,
  title = {{ProteoCraft: Protein Design as Proposal-Verification Search}},
  author = {Fang Wu and Hongbin Lin and Bingchen Duan and Yuchen Wang and Li Erran Li and Yanjie Huang and Mingxuan Cao and Yijun Ma and Yunxiang Mo and Yanchao Li and Kuan Pang and Juangui Xu and Hanqun Cao and Dongwei Jiang and Xinyue Xu and Xiangru Tang and Zehong Wang and Jian Ma and Kejun Ying and Guanlue Li and Mingran Cao and Fanjin Wang and Tianfan Fu and Shuangjia Zheng and Yao Shu and Jure Leskovec and Yejin Choi},
  howpublished = {arXiv},
  year = {2026},
  month = nov,
  doi = {10.48550/arxiv.2611.03963},
  eprint = {2611.03963},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2611.03963},
  note = {CC BY 4.0},
}

RIS

TY  - UNPB
TI  - ProteoCraft: Protein Design as Proposal-Verification Search
AU  - Fang Wu
AU  - Hongbin Lin
AU  - Bingchen Duan
AU  - Yuchen Wang
AU  - Li Erran Li
AU  - Yanjie Huang
AU  - Mingxuan Cao
AU  - Yijun Ma
AU  - Yunxiang Mo
AU  - Yanchao Li
AU  - Kuan Pang
AU  - Juangui Xu
AU  - Hanqun Cao
AU  - Dongwei Jiang
AU  - Xinyue Xu
AU  - Xiangru Tang
AU  - Zehong Wang
AU  - Jian Ma
AU  - Kejun Ying
AU  - Guanlue Li
AU  - Mingran Cao
AU  - Fanjin Wang
AU  - Tianfan Fu
AU  - Shuangjia Zheng
AU  - Yao Shu
AU  - Jure Leskovec
AU  - Yejin Choi
PY  - 2026
DA  - 2026/11/02
PB  - arXiv
DO  - 10.48550/arxiv.2611.03963
UR  - https://arxiv.org/abs/2611.03963
AB  - Protein-protein interactions underpin nearly all biological processes. Yet, designing proteins with desired interactions remains challenging due to the structural complexity governing molecular recognition. This study views protein design as a search optimization problem over a combinatorial sequence space under expensive structural evaluation. Toward this goal, we introduce a novel tree-based search framework, ProteoCraft, that employs approximate structural gradients from diffusion-based structure predictors as effective proposal policies to specify where the search should explore. Complete no-gradient inference of folding models separately assigns values to the resulting sequences, determining what is good. ProteoCraft allocates this verification budget through a coarse-to-fine paradigm: multi-site edits first explore distinct sequence basins, then verified promising states undergo strict single-site refinement. Across fourteen de novo binder-design settings spanning eight target proteins, five immune-relevant TCR-facing targets, and four localized CDR redesign tasks, ProteoCraft improves design quality and high-confidence yield. These improvements persist under independent AF2-IG and Rosetta-based interface evaluation. Prospective experimental validation further yielded an 82.9% binding hit rate for PD-L1 miniproteins, with a best measured KD of 0.879 nM, and confirmed binding of designed IFNAR2 VHH. Our code is available at https://github.com/smiles724/ProteoCraft_v1.
ER  -

Abstract

Related

Cite