We gave our new C2S-Scale 27B model a task: Find a drug that works like a conditioned reinforcerone that would boost the immune signal only in a specific “immune context positive” environment where low levels of interferon (a central immune signaling protein) were already present but insufficient to induce antigen presentation alone. This required a level of conditional reasoning that appeared to be a new scale ability; our smaller models could not resolve this context-dependent effect.
To achieve that, we have designed a dual context virtual screen to find this specific synergistic effect. The virtual screen involved two phases:
- Immune-context-positive: We fed the model with real-world patient samples with intact tumor-immune interactions and low-level interferon signaling.
- Immune-context-neutral: We fed the model with isolated cell line data without immune context.
We then simulated the effect of over 4,000 drugs across both contexts and asked the model to predict which drugs would only boost the antigen presentation in the first context, to bias the screen towards the patient-relevant setting. Out of the many drug candidates that the model highlights, a fraction (10-30%) of drug hits are already known in previous literature, while the remaining drugs are surprising hits without previously known link to the screen.
From prediction to experimental validation
The model’s predictions were clear. It identified a striking “context splitting” for the kinase CK2 inhibitor called silmitasertib (CX-4945). The model predicted a strong increase in antigen presentation when silmitasertib was applied in the “immune-context-positive” setting, but little or no effect in the “immune-context-neutral”. What made this prediction so exciting was that it was a novel idea. Although CK2 has been implicated in many cellular functions, including as a modulator of the immune system, inhibition of CK2 via silmitasertib has not been reported in the literature to explicitly increase MHC-I expression or antigen presentation. This highlights that the model generated a new, testable hypothesis and did not simply repeat known facts.
However, a prediction is only valuable if it can be validated in clinical use. The real test is first in the laboratory and finally in the clinic.
In the next phase of our project, we took this hypothesis to the lab bench and tested it in human neuroendocrine cell models—a cell type that was completely unseen by the model during training. The experiments showed:
- Treatment of the cells with silmitasertib alone had no effect on antigen presentation (MHC-I).
- Treatment of the cells with a low dose of interferon alone had a modest effect.
- Treatment of the cells with both silmitasertib and low-dose interferon produced a marked, synergistic amplification of antigen presentation.
Remarkably, in our laboratory tests, the combination of silmitasertib and low-dose interferon resulted in about a 50% increase in antigen presentation, which would make the tumor more visible to the immune system.
The model’s in silico the prediction was confirmed several times in vitro. C2S-Scale had successfully identified a novel interferon-dependent enhancer, revealing a new potential pathway to make “cold” tumors “hot” and potentially more responsive to immunotherapy. Although this is an early first step, it provides a powerful, experimentally validated lead for the development of new combination therapies that use multiple drugs together to achieve a more robust effect.
This result also provides a blueprint for a new kind of biological discovery. It shows that by following the scaling laws and building larger models like C2S-Scale 27B, we can create predictive models of cellular behavior that are powerful enough to drive high-throughput virtual displays, discover contextual biology, and generate biologically informed hypotheses.
Teams at Yale are now exploring the mechanism revealed here and further testing AI-generated predictions in other immune contexts. With further preclinical and clinical validation, such hypotheses may ultimately accelerate the path to new treatments.
Get started with the C2S-Scale 27B
The new C2S-Scale 27B model and its resources are available to the research community today. We invite you to explore these tools, build on our work, and help us continue to translate the language of life.