Google и Yale совместно создали ИИ для преобразования «холодных» опухолей в «горячие» Translation: Google and Yale Collaborate to Create AI for Transforming Cold Tumors into Hot Ones

Google, in collaboration with Yale University, has unveiled a new foundational model comprising 27 billion parameters, designed to decode the “language” of individual cells.

C2S-Scale 27B proposed a hypothesis regarding the behavior of cancer cells, which was subsequently validated through experimental trials on live biological samples.

“This discovery has revealed a promising avenue for creating new cancer therapy methods,” the company emphasized.

The model builds upon prior research that demonstrated that biological and linguistic systems adhere to similar scaling laws—where increased size equates to enhanced efficiency.

A major challenge in cancer immunotherapy is that many tumors remain «cold,» rendering them imperceptible to the immune system. One approach to «warm» them up involves triggering signal representation through a process known as «antigen presentation.»

At Google, a specific task was set for C2S-Scale 27B: to identify a drug that acts as a conditional amplifier, enhancing immune responses exclusively in a particular “immunopositive” environment where low levels of interferon already exist, but are insufficient for independently activating antigen presentation.

This task necessitated the use of conditional reasoning, a challenge that smaller models struggled to handle.

To achieve this objective, the researchers developed a virtual two-context screening system capable of identifying the desired synergistic effect. The process comprised two stages:

Subsequently, Google modeled over 4000 drugs across both contexts and tasked the model with pinpointing which ones amplified antigen presentation exclusively in the first context. This approach allowed for a focused search on clinically significant scenarios.

Among the numerous options, 10-30% were previously mentioned in scientific literature, while the remainder turned out to be unexpected discoveries.

The model uncovered a “striking contextual gap” for the CK2 kinase inhibitor silmitasertib (CX-4945). The neural network predicted a notable enhancement in antigen presentation when utilizing the drug in an «immunopositive» context, contrasted with almost no effect in an «immuno-neutral» environment.

What is particularly remarkable is that this represents a brand-new concept that had not been mentioned previously.

In the following phase, researchers tested the hypothesis in the laboratory, using human neuroendocrine cells—samples not encountered by the model during training. Results indicated that:

In laboratory experiments, this combination resulted in approximately a 50% increase in antigen presentation, making the tumor more detectable to the immune system.

The digitally made prediction was repeatedly validated.

C2S-Scale discovered a new conditional interferon amplifier that could assist in converting “cold” tumors into “hot” ones—more receptive to immunotherapy.

“Though this is merely the first step, it already provides an experimentally confirmed foundation for developing new combination therapies where multiple drugs work in concert for a more potent effect,” stated the blog.

Yale University’s teams are already investigating the identified mechanism and testing other AI predictions in various immune contexts. With further preclinical and clinical validation, such hypotheses could expedite the path toward new treatment methods.

Meanwhile, the biotechnology company SpotitEarly has begun developing a home cancer test based on analyzing human breath, combining canine olfaction with artificial intelligence algorithms.

Recall that in September, scientists developed an AI tool capable of predicting over 1000 diseases and forecasting health changes up to 10 years into the future.