
Researchers from Google DeepMind and Yale University have created an AI model that predicts a previously unknown mechanism of drug-immune system interaction, providing insights into cancer treatments.
The developed system, called Cell2Sentence-Scale 27B (C2S-Scale), is based on the Gemma family of open models with 27 billion parameters.
In the study, C2S-Scale simulated the effects of over 4,000 drugs on cancer cells and isolated the inhibitor silmitasertib. The model showed that the drug significantly enhanced antigen presentation (MHC-I) in an immune-reactive environment, making tumors more visible to the immune system.
Combining silmitasertib with low-dose interferon increased antigen reactivity by approximately 50%, an effect not seen with either drug alone. These results suggest the potential to transform so-called cold tumors, unrecognized by the immune system, into treatable tumors.