AI Model Reasons Through How Cells Change
PertReason, a research-stage system from Texas A&M, aims to explain how genes and drugs alter cells rather than only predicting disease.
Updated on Nov. 20, 2026 in Life Sciences

Texas A&M University researchers have developed PertReason, an AI virtual cell model that predicts how genetic changes and drugs alter cellular behavior. The work is research-stage: it has been described in an arXiv preprint and accepted for presentation at NeurIPS 2026, and is not an available product.
Why it matters
Current AI models can predict whether a person with a genetic change will develop a disease, but cannot explain why it occurred or how to design a treatment. PertReason aims to close that gap by reasoning through biological pathways, a direction that could shift AI in life sciences from prediction toward mechanistic explanation.
The evidence is a preprint describing PertReason and its PertReasonQA benchmark, which tests whether a model reaches correct answers through proper data rather than by chance. No measured accuracy figures or baseline comparisons were available, and the model's predictive performance remains unquantified in the source material.
The players
Texas A&M University
A public research university whose researchers are developing AI systems for biological modeling, and which recently brought the VISION supercomputer online on campus.
Shen
The Texas A&M lead researcher on PertReason, who plans to advance AI virtual cell models toward multiscale digital twins of the human body.
The details
PertReason traces how drugs and genetic changes affect genetic cellular networks, the interconnected sets of genes and proteins that govern how cells behave. The team applies electrical engineering principles to follow the processes behind cellular changes, treating a cell's signaling pathways somewhat like circuits whose signals can be traced. To check whether the model truly reasons, the researchers built PertReasonQA, a benchmark that flags cases where a model produced the right answer without using the proper underlying data. The system is trained to reason through biological pathways rather than only output a disease-risk score.
Timeline
July 2026: an earlier version of PertReason was presented at the 3rd Workshop on Multi-modal Foundation Models and Large Language Models for Life Sciences during ICML 2026 in Seoul, South Korea.
2026: PertReason was accepted for presentation at the Conference on Neural Information Processing Systems (NeurIPS 2026).
The Tech Race
PertReason sits among efforts to build AI virtual cell models, where the field has so far emphasized predicting disease outcomes from genetic data. Its emphasis on tracing mechanisms and benchmarking reasoning positions it as a different bet on what makes these models useful.
PertReason is a research-stage system described in a preprint and is not yet available for clinical or laboratory use, and no release date was announced. Its practical reach would come through scientists prioritizing biological pathways for study, but that use depends on validation not reported in the source.
The takeaway
The trajectory to watch is whether AI models can move from predicting disease outcomes to explaining the mechanisms behind them, which the NeurIPS 2026 presentation will put before the research community. Track the arXiv preprint and the PertReasonQA benchmark results for evidence on whether the reasoning holds up.
Further reading
For more on the biology behind this work, explore our Life Sciences coverage.
Source note: This article includes information reported by Phys.


