AI virtual cell uses protein dynamics to predict personalized breast cancer treatments

by Sanjukta Mondal, Medical Xpress

Sanjukta Mondal

Author

Meet our staff & contributors
Learn about our editorial standards

edited by Lisa Lock, reviewed by Robert Egan

Lisa Lock

Scientific Editor

Meet our editorial team
Behind our editorial process

Robert Egan

Senior Editor

Meet our editorial team
Behind our editorial process Editors' notes

This article has been reviewed according to Science X's editorial process and policies. Editors have highlighted the following attributes while ensuring the content's credibility:

fact-checked

peer-reviewed publication

trusted source

proofread

The GIST Add as preferred source

Not all cancer cells are the same, and neither are their responses to drugs. Finding the right drug for the right cancer cell is a long battle in drug development and in designing effective treatments. A new AI tool called ProteinTalks makes this task easier by accurately predicting whether a drug will be effective against a cancer cell line, finding new drug combinations and identifying proteins linked to drug resistance.

The team captured more than 38 million protein measurements to map how breast cancer cells respond to 63 FDA-approved anticancer drugs and 59 two-drug combinations. Instead of looking at a single moment, they followed the response over time, measuring before treatment and again at 6, 24 and 48 hours. This time-resolved dataset is what helps ProteinTalks learn how protein levels shift as the drug takes effect.

According to the report published in Nature, ProteinTalks outperformed existing AI models based on gene activity and traditional machine-learning methods in predicting how cancer cells respond to drugs. It accurately predicted responses to 81 anticancer compounds it had never seen during training and revealed four drug pairs that showed a stronger effect together than either drug alone against triple-negative breast cancer—an aggressive cancer that is hard to treat because of a lack of hormone receptors.

Laboratory in a chip

When testing new cancer treatments, scientists need to simulate how cells react to drugs with a computer program before running expensive lab trials. Virtual cells could help solve this problem if only researchers could find a way to recreate the nonlinear, complex changes that happen inside a cell over time.

Foundational AI models trained on large-scale omics data, which capture biological molecules such as DNA, RNA, proteins and metabolites, are opening a promising path toward an AI virtual cell. However, most models still see cell behavior as a snapshot rather than something that changes over time.

Proteins drive a cell's behavior and response to drugs, so tracking how their levels change over time can reveal how drugs work and how cells develop resistance. In this study, researchers treated 18 breast cancer cell lines with 63 approved drugs and a few dozen drug combinations, measuring protein levels and cell survival before and after treatment.

This produced more than 38 million protein measurements across 5,585 proteins, alongside thousands of cell survival data points. ProteinTalks was then trained on these protein-response trajectories alongside drug information to learn how cells respond to treatment over time.

What proteins reveal

They found that ProteinTalks functioned as an AI virtual cell that accurately simulated how cancer cells respond dynamically to individual anticancer compounds over time. Even though it was trained on breast cancer cell lines, it successfully predicted cell responses to lung, colorectal, pancreatic and melanoma cancer cells. By tracking protein changes over time, the model pinpointed key drivers of drug resistance. One example was AKR1C3, where knocking down the protein restored cancer cells' sensitivity to docetaxel, a chemotherapy medication, a finding confirmed through cytotoxicity assays.

The platform analyzed the unique protein signatures in 501 triple-negative breast cancer tumors and grouped patients by their treatment plans, which were based on the risk of recurrence and long-term survival outcomes.

By screening more than 3,000 repurposed drugs directly against patient-derived organoids, which are mini-tumors that mimic patients' cancers, ProteinTalks pinpointed personalized drug candidates that could kill tumor cells at far lower doses than standard chemotherapy. The researchers also obtained a new list of promising drug combinations that work synergistically in the fight against cancer.

Cancer treatment remains uncertain because a drug that works for one patient may not work for another. This new protein-based model, shaped by dynamic changes in cells over time, could be a turning point in the development of interpretable AI virtual cells that can pinpoint promising drugs before costly lab testing and guide personalized treatment based on tumor protein profiles.

Written for you by our author Sanjukta Mondal, edited by Lisa Lock, and fact-checked and reviewed by Robert Egan—this article is the result of careful human work. We rely on readers like you to keep independent science journalism alive. If this reporting matters to you, please consider a donation (especially monthly). You'll get an ad-free account as a thank-you.

Publication details

Rui Sun et al, An operational perturbation proteomics-based virtual cell model, Nature (2026). DOI: 10.1038/s41586-026-11001-9

Journal information: Nature

Key medical concepts

Triple Negative Breast NeoplasmsDrug Resistance

Clinical categories

OncologyClinical pharmacology Who's behind this story?

Sanjukta Mondal

Master's in Chemistry. Freelance science journalist and communicator. Published in Chemistry World, BioSpace, and The Hindu. Full profile →

Lisa Lock

BA art history, MA material culture. Former museum editor, paramedic, and transplant coordinator. Editing for Science X since 2021. Full profile →

Robert Egan

Bachelor's in mathematical biology, Master's in creative writing. Well-traveled with unique perspectives on science and language. Full profile →

© 2026 Science X Network

Citation: AI virtual cell uses protein dynamics to predict personalized breast cancer treatments (2026, September 12) retrieved 12 September 2026 from https://medicalxpress.com/news/2026-09-ai-virtual-cell-protein-dynamics.html This document is subject to copyright. Apart from any fair dealing for the purpose of private study or research, no part may be reproduced without the written permission. The content is provided for information purposes only.