MechPert: Mechanistic Consensus as an Inductive Bias for Unseen Perturbation Prediction

TL;DR

MechPert introduces a novel framework using mechanistic consensus among language model agents to improve prediction of gene responses to unseen perturbations, outperforming existing methods especially in low-data scenarios.

Contribution

It proposes a new mechanistic consensus approach that leverages directed regulatory hypotheses from LLMs for better perturbation prediction.

Findings

Improves Pearson correlation by up to 10.5% in low-data regimes.

Outperforms similarity-based baselines and network heuristics in gene perturbation prediction.

Effective across multiple human cell lines.

Abstract

Predicting transcriptional responses to unseen genetic perturbations is essential for understanding gene regulation and prioritizing large-scale perturbation experiments. Existing approaches either rely on static, potentially incomplete knowledge graphs, or prompt language models for functionally similar genes, retrieving associations shaped by symmetric co-occurrence in scientific text rather than directed regulatory logic. We introduce MechPert, a lightweight framework that encourages LLM agents to generate directed regulatory hypotheses rather than relying solely on functional similarity. Multiple agents independently propose candidate regulators with associated confidence scores; these are aggregated through a consensus mechanism that filters spurious associations, producing weighted neighborhoods for downstream prediction. We evaluate MechPert on Perturb-seq benchmarks across four human cell lines. For perturbation prediction in low-data regimes ($N=50$ observed perturbations), MechPert improves Pearson correlation by up to 10.5\% over similarity-based baselines. For experimental design, MechPert-selected anchor genes outperform standard network centrality heuristics by up to 46\% in well-characterized cell lines.