Progress over time
A virtual cell model learns how cancer cells respond to drugs over time
ProteinTalks learned from more than 38 million protein measurements taken after breast-cancer cells were exposed to drugs. It predicted responses in laboratory datasets and helped rank candidates for further testing. The work does not show that using the model improves treatment for patients.
Sun and colleagues · Nature · September 9, 2026
Research claim
AI-designed molecules survived a first laboratory test against a tuberculosis target
A chemical language model proposed molecules for a tuberculosis protein, and laboratory tests found 14 generated or closely related compounds that inhibited the target. They are early research hits, not medicines, and their safety and activity in living systems remain unknown.
Wu and colleagues · Nature Communications · October 29, 2024
Independent validation
2 of 2 results. End of results.