PertReason: A Knowledge-Grounded Benchmark and Framework for Cell-State-Conditioned Mechanistic Reasoning of Perturbation Effects
Abstract
Evaluating machine learning in scientific domains requires separating correct predictions from correct reasons under realistic distribution shifts. We introduce PertReason, a knowledge-grounded approach to evaluating and improving cell-state-conditioned reasoning about perturbation effects. Our framework comprises PertReasonQA, a benchmark that provides evaluation and supervision, and PertReasonLM, a reference large language model that demonstrates how such supervision supports reasoning-centric training. PertReasonQA tests whether models can generate mechanistically faithful explanations while remaining robust to complex shifts, such as new cells, unseen perturbations, cross-modal or combinatorial extrapolations. PertReasonQA combines single-cell genetic and chemical perturbation data across multiple cellular contexts with knowledge graphs, and dynamically conditions pathways on cell-specific basal states to avoid generic memorization. Evaluations of state-of-the-art models reveal systematic gaps between outcome prediction and mechanistic reasoning. As a reference probe of the benchmark, PertReasonLM is trained through supervised fine-tuning and reinforcement learning using mechanistic supervision from PertReasonQA. PertReasonLM improves predictive accuracy and reduces answer-reason disagreement by grounding rationales in context-specific pathways. Together, we provide a diagnostic framework for exposing and mitigating failures in faithful reasoning in data-rich scientific systems. We release our datasets to support further research at https://dongkwan-kim.github.io/PertReason/.
Related
- A rank-conditioned α-stable perturbation for differential evolution: Separating perturbation frequency and tail heaviness
- Where single-cell foundation models help: a benchmark of annotation, perturbation and regulatory inference against simple baselines
- AnnFlux: object-conditioned neural stochastic differential equations for single-cell perturbation dynamics
- Centered cross-context perturbation transfer on the public 2025 Virtual Cell Challenge benchmark
- Genetic Perturbation Prediction Models Fail Out-of-Distribution: A Rigorous Benchmark and Systematic Evaluation
BibTeX
@misc{kim2026pertreason,
title = {{PertReason: A Knowledge-Grounded Benchmark and Framework for Cell-State-Conditioned Mechanistic Reasoning of Perturbation Effects}},
author = {Dongkwan Kim and Yiming Gao and Yining Yang and Yang Shen},
howpublished = {arXiv},
year = {2026},
month = nov,
doi = {10.48550/arxiv.2607.18777},
eprint = {2607.18777},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2607.18777},
}
RIS
TY - UNPB
TI - PertReason: A Knowledge-Grounded Benchmark and Framework for Cell-State-Conditioned Mechanistic Reasoning of Perturbation Effects
AU - Dongkwan Kim
AU - Yiming Gao
AU - Yining Yang
AU - Yang Shen
PY - 2026
DA - 2026/11/11
PB - arXiv
DO - 10.48550/arxiv.2607.18777
UR - https://arxiv.org/abs/2607.18777
AB - Evaluating machine learning in scientific domains requires separating correct predictions from correct reasons under realistic distribution shifts. We introduce PertReason, a knowledge-grounded approach to evaluating and improving cell-state-conditioned reasoning about perturbation effects. Our framework comprises PertReasonQA, a benchmark that provides evaluation and supervision, and PertReasonLM, a reference large language model that demonstrates how such supervision supports reasoning-centric training. PertReasonQA tests whether models can generate mechanistically faithful explanations while remaining robust to complex shifts, such as new cells, unseen perturbations, cross-modal or combinatorial extrapolations. PertReasonQA combines single-cell genetic and chemical perturbation data across multiple cellular contexts with knowledge graphs, and dynamically conditions pathways on cell-specific basal states to avoid generic memorization. Evaluations of state-of-the-art models reveal systematic gaps between outcome prediction and mechanistic reasoning. As a reference probe of the benchmark, PertReasonLM is trained through supervised fine-tuning and reinforcement learning using mechanistic supervision from PertReasonQA. PertReasonLM improves predictive accuracy and reduces answer-reason disagreement by grounding rationales in context-specific pathways. Together, we provide a diagnostic framework for exposing and mitigating failures in faithful reasoning in data-rich scientific systems. We release our datasets to support further research at https://dongkwan-kim.github.io/PertReason/.
ER -