AI and science, promising advances

Discovering new cancer therapies has historically been an immense challenge…

Beyond LLMs...

Discovering new cancer therapies has historically been an immense challenge. It has relied on compound screening processes that are expensive and often hit-or-miss.

Today that approach is starting to change, thanks to artificial intelligence applied to biology. AI models can now analyze biological data at a scale that was out of our reach, identifying correlations and mechanisms of action that conventional human analysis cannot detect. That could speed up the discovery pipeline dramatically. One example is C2S-Scale 27B, introduced by researchers at Google and Yale University. It is a foundation model (based on the Gemma family) that identified a path to cancer treatment that had been unknown until now.

The process the model followed is fascinating:

· The system processes complex cellular data (such as gene expression) as if it were a language, which lets it model how cells behave and respond to different interventions.

· The model was instructed to find compounds that would raise the "visibility" of tumor cells to the immune system, but only under specific molecular conditions.

· The AI identified silmitasertib, an existing drug with no previously known link to this mechanism of immune recognition. In vitro validation confirmed what the AI predicted: the drug combination increased the visibility of tumor cells to immune defenses by 50%.

In other words, artificial intelligence is shifting from an optimization tool to a real partner in scientific discovery. We see it in its ability to generate novel hypotheses, validated empirically, that could redefine the pace of biomedical innovation.

There is life beyond generative AI...

The announcement: https://blog.google/technology/ai/google-gemma-ai-cancer-therapy-discovery/