A team at Westlake University in Hangzhou has built a virtual-cell model that predicts how a person’s triple-negative breast cancer will respond to treatment, working from protein measurements rather than gene sequencing.
The training set is unusually large for the field. It spans 18 breast cancer cell lines, most of them TNBC, and more than 38 million protein measurements.
Cells met 63 FDA-approved cancer drugs plus 59 combinations. Protein levels were sampled at four points: baseline, then 6, 24 and 48 hours. The time course mattered, because static snapshots miss how cells change.
Arc Institute systems biologist Hani Goodarzi, who was not involved in the study, said repeated sampling is what separates this work from most virtual-cell efforts.
On 81 drugs left out of training, the tool hit 88% accuracy. It also read 3,651 proteins in pre-chemotherapy biopsies from 501 patients and matched the outcomes they later had.
TNBC accounts for 15-20% of breast cancer cases and lacks the receptors that hormone and targeted therapies need, so treatment selection leans heavily on trial and error. Co-author Tiannan Guo called this the first time a virtual-cell model has been tested in a clinical scenario.
