Rank-and-Reason: Multi-Agent Collaboration Accelerates Zero-Shot Protein Mutation Prediction

2026-01-30

AI Summary

This work addresses the limitations of current protein language models in zero-shot mutation prediction, which often ignore biophysical constraints and rely on inefficient, subjective manual curation. The authors propose Rank-and-Reason (VenusRAR), the first multi-agent collaborative two-stage framework: the first stage employs a context-aware, multimodal ensemble ranking to prioritize candidate mutations, while the second stage deploys a virtual expert panel that conducts chain-of-thought reasoning incorporating geometric and structural constraints for rigorous evaluation. Evaluated on ProteinGym, VenusRAR achieves a Spearman correlation of 0.551 and a 367% improvement in Top-5 hit rate. Wet-lab validation on Cas12i3 mutants yields a 46.7% positive rate, successfully identifying two novel variants with 4.23× and 5.05× enhanced activity, substantially improving both prediction reliability and experimental efficiency.

Technology Category

• Search and Optimization: Metareasoning and Metaheuristics
• Machine Learning: Large Multimodal Models (LMMs)
• Intelligent Robots: Multimodal Perception & Sensor Fusion

Application Category

• Search and Retrieval-Augmented AI: Web search models and ranking
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

Zero-shot mutation prediction is vital for low-resource protein engineering, yet existing protein language models (PLMs) often yield statistically confident results that ignore fundamental biophysical constraints. Currently, selecting candidates for wet-lab validation relies on manual expert auditing of PLM outputs, a process that is inefficient, subjective, and highly dependent on domain expertise. To address this, we propose Rank-and-Reason (VenusRAR), a two-stage agentic framework to automate this workflow and maximize expected wet-lab fitness. In the Rank-Stage, a Computational Expert and Virtual Biologist aggregate a context-aware multi-modal ensemble, establishing a new Spearman correlation record of 0.551 (vs. 0.518) on ProteinGym. In the Reason-Stage, an agentic Expert Panel employs chain-of-thought reasoning to audit candidates against geometric and structural constraints, improving the Top-5 Hit Rate by up to 367% on ProteinGym-DMS99. The wet-lab validation on Cas12i3 nuclease further confirms the framework's efficacy, achieving a 46.7% positive rate and identifying two novel mutants with 4.23-fold and 5.05-fold activity improvements. Code and datasets are released on GitHub (https://github.com/ai4protein/VenusRAR/).

Problem

• zero-shot protein mutation prediction
• protein language models
• biophysical constraints
• wet-lab validation
• expert auditing

Innovation

• multi-agent collaboration
• zero-shot protein mutation prediction
• chain-of-thought reasoning
• context-aware multimodal ensemble
• biophysical constraint auditing