Towards Autonomous Mechanistic Reasoning in Virtual Cells

Hugging Face Daily Papers Papers

Summary

This paper introduces VCR-Agent, a multi-agent framework that enhances large language models for biological research by generating and validating mechanistic explanations using structured formalism and the VC-TRACES dataset. The approach improves factual precision in gene expression prediction through verified mechanistic reasoning in virtual cells.

Large language models (LLMs) have recently gained significant attention as a promising approach to accelerate scientific discovery. However, their application in open-ended scientific domains such as biology remains limited, primarily due to the lack of factually grounded and actionable explanations. To address this, we introduce a structured explanation formalism for virtual cells that represents biological reasoning as mechanistic action graphs, enabling systematic verification and falsification. Building upon this, we propose VCR-Agent, a multi-agent framework that integrates biologically grounded knowledge retrieval with a verifier-based filtering approach to generate and validate mechanistic reasoning autonomously. Using this framework, we release VC-TRACES dataset, which consists of verified mechanistic explanations derived from the Tahoe-100M atlas. Empirically, we demonstrate that training with these explanations improves factual precision and provides a more effective supervision signal for downstream gene expression prediction. These results underscore the importance of reliable mechanistic reasoning for virtual cells, achieved through the synergy of multi-agent and rigorous verification.
Original Article
View Cached Full Text

Cached at: 04/20/26, 08:28 AM

Paper page - Towards Autonomous Mechanistic Reasoning in Virtual Cells

Source: https://huggingface.co/papers/2604.11661

Abstract

Large language models are enhanced for biological research through a multi-agent framework that generates and validates mechanistic explanations using structured formalism and verified datasets.

Large language models (https://huggingface.co/papers?q=Large%20language%20models)(LLMs) have recently gained significant attention as a promising approach to accelerate scientific discovery. However, their application in open-ended scientific domains such as biology remains limited, primarily due to the lack of factually grounded and actionable explanations. To address this, we introduce a structured explanation formalism forvirtual cells (https://huggingface.co/papers?q=virtual%20cells)that represents biological reasoning asmechanistic action graphs (https://huggingface.co/papers?q=mechanistic%20action%20graphs), enabling systematic verification and falsification. Building upon this, we propose VCR-Agent, amulti-agent framework (https://huggingface.co/papers?q=multi-agent%20framework)that integratesbiologically grounded knowledge retrieval (https://huggingface.co/papers?q=biologically%20grounded%20knowledge%20retrieval)with averifier-based filtering (https://huggingface.co/papers?q=verifier-based%20filtering)approach to generate and validatemechanistic reasoning (https://huggingface.co/papers?q=mechanistic%20reasoning)autonomously. Using this framework, we releaseVC-TRACES dataset (https://huggingface.co/papers?q=VC-TRACES%20dataset), which consists of verified mechanistic explanations derived from theTahoe-100M atlas (https://huggingface.co/papers?q=Tahoe-100M%20atlas). Empirically, we demonstrate that training with these explanations improves factual precision and provides a more effective supervision signal for downstreamgene expression prediction (https://huggingface.co/papers?q=gene%20expression%20prediction). These results underscore the importance of reliablemechanistic reasoning (https://huggingface.co/papers?q=mechanistic%20reasoning)forvirtual cells (https://huggingface.co/papers?q=virtual%20cells), achieved through the synergy of multi-agent and rigorous verification.

View arXiv page (https://arxiv.org/abs/2604.11661)View PDF (https://arxiv.org/pdf/2604.11661)Project page (https://valencelabs.substack.com/p/towards-reasoning-in-virtual-cells)GitHub4 (https://github.com/valence-labs/VCR-Agent)Add to collection (https://huggingface.co/login?next=%2Fpapers%2F2604.11661)

Get this paper in your agent:

hf papers read 2604\.11661

Don’t have the latest CLI?curl \-LsSf https://hf\.co/cli/install\.sh \| bash

Models citing this paper0

No model linking this paper

Cite arxiv.org/abs/2604.11661 in a model README.md to link it from this page.

Datasets citing this paper0

No dataset linking this paper

Cite arxiv.org/abs/2604.11661 in a dataset README.md to link it from this page.

Spaces citing this paper0

No Space linking this paper

Cite arxiv.org/abs/2604.11661 in a Space README.md to link it from this page.

Collections including this paper0

No Collection including this paper

Add this paper to acollection (https://huggingface.co/new-collection)to link it from this page.

Similar Articles

Inside VAKRA: Reasoning, Tool Use, and Failure Modes of Agents

Hugging Face Blog

This article introduces VAKRA, an executable benchmark for evaluating AI agents' reasoning and tool-use capabilities in enterprise-like environments. It analyzes failure modes and details the benchmark's structure involving API chaining and document retrieval.

From Residuals to Reasons: LLM-Guided Mechanism Inference from Tabular Data

arXiv cs.LG

Introduces Multi-Agent Residual In-Context Learning (MARICL), an agentic framework that uses LLM agents to analyze residuals from a base model on tabular data, hypothesize missing structure, and produce explicit correction terms via textual gradient optimization. Across nine benchmarks, MARICL consistently improves over its base model and demonstrates mechanistic generalization in cell-free protein predictions.

AgentV-RL: Scaling Reward Modeling with Agentic Verifier

arXiv cs.CL

AgentV-RL introduces an Agentic Verifier framework that enhances reward modeling through bidirectional verification with forward and backward agents augmented with tools, achieving 25.2% improvement over state-of-the-art ORMs. The approach addresses error propagation and grounding issues in verifiers for complex reasoning tasks through multi-turn deliberative processes combined with reinforcement learning.