Chain of Structures: Connecting Reasoning and Generation for Protein Design
- 1HKUST
Abstract
Scientific design requires reasoning in scientific spaces that represent the objects being designed. Building on Scinetics' Chain of Science Tokens, we present chain-of-structures (COS) as a concrete protein-design case study that connects reasoning and generation through explicit structural states. In this technical report, we apply COS to enzyme design, where catalytic function requires precise local geometry within a compatible global fold, making it a demanding test of reasoning from chemical requirements to protein structure. Starting from a reaction specification and target length, COS couples a language reasoning expert with an all-atom generation expert across four rounds: catalytic core, active site, global backbone, and complete sequence and structure. Each reasoning state directly conditions generation, and the resulting structure becomes context for the next decision. Discrete residue constraints and continuous latents connect the two experts, enabling structural supervision of the reasoning interface. On the evaluated benchmark, COS leads the compared baselines on five of six metrics, achieving 57.0% EC3 agreement and a TM-score of 0.729.
Related
- Evidence Scaling for Zero-Shot Protein Reasoning with Large Language Models
- TRACEDD: A Tool-grounded Reasoning and Agentic Coordination for Explainable Drug Design
- WGG: a graphical tool for visualizing genome structures using SNPs detected by next-generation sequencing.
- TopoFlow: Evolutionarily Conditioned Flow Matching for Protein Conformational Ensemble Generation
- Suramin inspires next-generation broad-spectrum antiviral design targeting RdRps
BibTeX
@misc{cai2026chain,
title = {{Chain of Structures: Connecting Reasoning and Generation for Protein Design}},
author = {Q. Cai and M. Xu and J. Yan and Z. Wang and Y. Liu and Q. Pei and Z. Zhang},
howpublished = {bioRxiv},
year = {2026},
month = oct,
doi = {10.64898/2026.10.09.758133},
url = {https://www.biorxiv.org/content/10.64898/2026.10.09.758133v1},
note = {CC BY},
}
RIS
TY - UNPB
TI - Chain of Structures: Connecting Reasoning and Generation for Protein Design
AU - Q. Cai
AU - M. Xu
AU - J. Yan
AU - Z. Wang
AU - Y. Liu
AU - Q. Pei
AU - Z. Zhang
PY - 2026
DA - 2026/10/09
PB - bioRxiv
DO - 10.64898/2026.10.09.758133
UR - https://www.biorxiv.org/content/10.64898/2026.10.09.758133v1
AB - Scientific design requires reasoning in scientific spaces that represent the objects being designed. Building on Scinetics' Chain of Science Tokens, we present chain-of-structures (COS) as a concrete protein-design case study that connects reasoning and generation through explicit structural states. In this technical report, we apply COS to enzyme design, where catalytic function requires precise local geometry within a compatible global fold, making it a demanding test of reasoning from chemical requirements to protein structure. Starting from a reaction specification and target length, COS couples a language reasoning expert with an all-atom generation expert across four rounds: catalytic core, active site, global backbone, and complete sequence and structure. Each reasoning state directly conditions generation, and the resulting structure becomes context for the next decision. Discrete residue constraints and continuous latents connect the two experts, enabling structural supervision of the reasoning interface. On the evaluated benchmark, COS leads the compared baselines on five of six metrics, achieving 57.0% EC3 agreement and a TM-score of 0.729.
ER -