Salvato in:
Dettagli Bibliografici
Autori principali: Martell, Marc Boubnovski, Stoisser, Josefa Lia, Phillips, Lawrence, Misra, Aditya, Kitchen, Robert, Ferkinghoff-Borg, Jesper, Yu, Jialin, Torr, Philip, Märten, Kaspar
Natura: Preprint
Pubblicazione: 2026
Soggetti:
Accesso online:https://arxiv.org/abs/2602.13791
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866915798512041984
author Martell, Marc Boubnovski
Stoisser, Josefa Lia
Phillips, Lawrence
Misra, Aditya
Kitchen, Robert
Ferkinghoff-Borg, Jesper
Yu, Jialin
Torr, Philip
Märten, Kaspar
author_facet Martell, Marc Boubnovski
Stoisser, Josefa Lia
Phillips, Lawrence
Misra, Aditya
Kitchen, Robert
Ferkinghoff-Borg, Jesper
Yu, Jialin
Torr, Philip
Märten, Kaspar
contents 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.
format Preprint
id arxiv_https___arxiv_org_abs_2602_13791
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle MechPert: Mechanistic Consensus as an Inductive Bias for Unseen Perturbation Prediction
Martell, Marc Boubnovski
Stoisser, Josefa Lia
Phillips, Lawrence
Misra, Aditya
Kitchen, Robert
Ferkinghoff-Borg, Jesper
Yu, Jialin
Torr, Philip
Märten, Kaspar
Machine Learning
Artificial Intelligence
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.
title MechPert: Mechanistic Consensus as an Inductive Bias for Unseen Perturbation Prediction
topic Machine Learning
Artificial Intelligence
url https://arxiv.org/abs/2602.13791