Google DeepMind’s artificial intelligence has made a breakthrough in finding new methods for cancer treatment through the Cell2Sentence-Scale biomodel, which contains 27 billion parameters. This model not only analyzes data at the individual cell level but also independently formulates and experimentally tests scientific hypotheses.
This is reported by Finway
A New Approach to Enhancing Immune Response
In collaboration with Yale University, the DeepMind team utilized the Cell2Sentence-Scale (C2S-Scale) model, built on the Gemma architecture. The main achievement is the identification of the drug silmitasertib (CX-4945), which can enhance the immune visibility of tumor cells. This drug allows the immune system to better recognize and respond to tumors, opening a new direction for developing effective therapeutic methods.
To validate the hypothesis, C2S-Scale analyzed the effects of over 4,000 different substances under conditions of active immune signaling. The results showed that silmitasertib significantly increases antigen presentation—a key process that activates the immune response. Importantly, this effect is observed only under conditions of an active immune system.
Experimental Confirmations and Prospects
Laboratory studies on human cells confirmed the model’s hypothesis: when silmitasertib was combined with a low dose of interferon, the level of antigen presentation increased by 50%. This is the first instance where artificial intelligence has proposed a new, previously unreported combination with significant clinical potential.
“DeepMind notes that scaling biomodels not only allows for more accurate predictions but also generates fundamentally new ideas.”
Researchers at Yale University are currently continuing to explore the mechanisms of the identified effect and are testing other predictions of the system. According to Google CEO Sundar Pichai, this discovery could serve as a foundation for new clinical research and enhances the role of artificial intelligence in fundamental biological research. All developments, including code and models, are already available on the Hugging Face and GitHub platforms, and the preprint of the study has been published on bioRxiv.
At the same time, experts emphasize that the obtained results have not yet undergone the peer review process, thus requiring further verification and confirmation of efficacy for practical application in therapy.