Researchers at Texas A&M University have built an AI virtual cell model that predicts how genetic changes and drugs alter the behaviour of cells, and traces the biological pathways behind those changes. The system, called PertReason, was developed in the university’s Department of Electrical and Computer Engineering. It has been accepted for presentation at the Conference on Neural Information Processing Systems, known as NeurIPS 2026, and the paper is available on the arXiv preprint server.
Today’s AI models can predict whether a person with a genetic change will develop a disease, but they cannot answer two questions that matter to doctors: why the disease occurred, and how to design a treatment. The virtual cell project aims to close that gap. Yang Shen, a professor of electrical and computer engineering, is working on the project with postdoctoral researcher Dongkwan Kim and doctoral student Yiming Gao, who lead the work. Doctoral student Yining Yang is also on the team.
Many traditional methods focus on single biological components. The virtual cell approach traces how drugs and genetic changes affect genetic cellular networks instead. That wider view helps researchers understand why a biological change leads to disease and where to look for a treatment. “In the past, much of our work focused on fundamental molecules of life, such as proteins,” Shen said. “Now, our work extends beyond the molecular level to the cellular level. We’re expanding the scale of what we study.”
Virtual cell model built to explain its answers
The virtual cell team applied electrical engineering principles to biology, using the approach to uncover the processes behind cellular changes. They also trained an AI model to reason through biological pathways, which they say makes its predictions more trustworthy. A benchmark called PertReasonQA tests more than whether a language model gets the right answer: it checks whether the reasoning behind the answer is accurate, and it catches cases where a model produced a correct answer without using the proper data.
An earlier version of the work was presented in July at the 3rd Workshop on Multi-modal Foundation Models and Large Language Models for Life Sciences, held during the International Conference on Machine Learning, known as ICML 2026, in Seoul, South Korea. The virtual cell paper is listed on arXiv as “PertReason: A Knowledge-Grounded Benchmark and Framework for Cell-State-Conditioned Mechanistic Reasoning of Perturbation Effects”, by Dongkwan Kim and colleagues.
Virtual cell research shaped by breast cancer cases
Early in the research, Shen and his team began working with clinical scientists treating real breast cancer patients. They saw the problems patients face, above all those who respond to treatment at first and then suddenly develop resistance. “It’s one thing if your AI model has nice numerical metrics, which demonstrate technical success. But it’s another thing if your work can actually make an impact on people,” Shen said. “We know that genetic changes might be responsible for cancer metastasis, but the natural question is why it starts and how it can be addressed.”
By helping scientists narrow down which biological pathways to investigate first, the virtual cell system could cut trial and error and put the most promising pathways at the front of the queue, saving time and resources. The team says the work could lead to a deeper understanding of disease pathways, help identify new drug targets and support the design of better treatments.
The next step is scale. Texas A&M’s supercomputer VISION recently arrived on campus, and Shen plans to use it to advance the virtual cell work toward virtual replicas of the human body, known as digital twins, that could inform medical decisions. The work builds on AI virtual cell models, and Shen said the aim is to move from single cells to whole patients. “I’m looking forward to accomplishing our vision of creating multiscale digital twins for patients and developing personalized medicine,” Shen said. “The goal is always to support clinicians and improve patient outcomes.”
The team continues to work on the model following its acceptance to NeurIPS, with the long-term aim of using AI to find new ways to understand and treat disease for each patient. For now, PertReason remains a research tool, tested on data rather than in clinics, but its authors say the reasoning it shows is the step that matters.

