Accurate and interpretable protein complex stoichiometry prediction through agentic AI with StoiAgent

Accurate protein complex structure modeling requires knowledge of complex stoichiometry, yet the copy numbers of individual subunits are often unknown. Consequently, predicting protein complex stoichiometry has emerged as a major challenge in recent years. Here, we present StoiAgent, a large language model (LLM)-powered AI agent that automates the retrieval and integration of sequence, structural, and literature evidence to predict stoichiometry while preserving interpretable reasoning traces (chain of thoughts) of its evidence assessment and ranking decision. An open-weight LLM model, Qwen3.6-27B, is used by StoiAgent as the LLM backbone to combine sequences, structural templates, AlphaFold3 structure predictions, and literature evidence, and perform stoichiometry candidate generation and ranking. On 56 CASP17 protein complex targets, it achieves top-1 accuracy of 75.0%, which represents a 23.2 percentage-point improvement over PreStoi with AlphaFold3 ranking (51.8%) and matches the top-1 accuracy of StoPred. Moreover, StoiAgent has a higher top-3 and top-5 accuracy than StoPred (85.7% versus 80.4% and 87.5% versus 85.7%, respectively). Furthermore, it outperforms general-purpose frontier LLMs, including ChatGPT and Gemini, in stoichiometry prediction. Case studies further illustrate how StoiAgent evaluates competing assembly hypotheses by integrating complementary evidence, with recorded reasoning traces documenting its decision-making process. Overall, StoiAgent demonstrates the potential of LLM-powered AI agents to improve protein complex stoichiometry prediction through the automated integration of diverse sources of evidence while providing interpretable explanations to guide downstream structure modeling.

Authors

Institutions

Publication Details

Journal
bioRxiv (Cold Spring Harbor Laboratory)
Published
2026-11-18
DOI
https://doi.org/10.64898/2026.11.17.767325
Primary Topic
Protein Structure and Dynamics
Type
preprint

Funders

Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

Accurate and interpretable protein complex stoichiometry prediction through agentic AI with StoiAgent

Jian Liu, Jianlin Jack Cheng, Pawan Neupane
2 links
bioRxiv (Cold Spring Harbor Laboratory)
Protein Structure and Dynamics
preprint

Accurate and interpretable protein complex stoichiometry prediction through agentic AI with StoiAgent

Jian Liu, Jianlin Jack Cheng, Pawan Neupane
preprint en

Abstract

Accurate protein complex structure modeling requires knowledge of complex stoichiometry, yet the copy numbers of individual subunits are often unknown. Consequently, predicting protein complex stoichiometry has emerged as a major challenge in recent years. Here, we present StoiAgent, a large language model (LLM)-powered AI agent that automates the retrieval and integration of sequence, structural, and literature evidence to predict stoichiometry while preserving interpretable reasoning traces (chain of thoughts) of its evidence assessment and ranking decision. An open-weight LLM model, Qwen3.6-27B, is used by StoiAgent as the LLM backbone to combine sequences, structural templates, AlphaFold3 structure predictions, and literature evidence, and perform stoichiometry candidate generation and ranking. On 56 CASP17 protein complex targets, it achieves top-1 accuracy of 75.0%, which represents a 23.2 percentage-point improvement over PreStoi with AlphaFold3 ranking (51.8%) and matches the top-1 accuracy of StoPred. Moreover, StoiAgent has a higher top-3 and top-5 accuracy than StoPred (85.7% versus 80.4% and 87.5% versus 85.7%, respectively). Furthermore, it outperforms general-purpose frontier LLMs, including ChatGPT and Gemini, in stoichiometry prediction. Case studies further illustrate how StoiAgent evaluates competing assembly hypotheses by integrating complementary evidence, with recorded reasoning traces documenting its decision-making process. Overall, StoiAgent demonstrates the potential of LLM-powered AI agents to improve protein complex stoichiometry prediction through the automated integration of diverse sources of evidence while providing interpretable explanations to guide downstream structure modeling.

bioRxiv (Cold Spring Harbor Laboratory)
University of Missouri (US)
National Institutes of Health, National Institute of General Medical Sciences
Protein Structure and Dynamics
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon