Particle.news
Download on the App Store
3 ARTICLES
·
2mo ago

DeepMind–Yale AI Identifies Drug Candidate That Raises Tumor Antigen Signals by 50% in Lab Tests

Using single‑cell modeling plus a virtual screen of thousands of compounds, the system surfaced a context‑dependent amplifier that could help make 'cold' tumors more visible to the immune system.

A lab technician works on a research process to find new CAR-T cells and RNA in the laboratory of French biopharmaceutical company Cellectis in Paris on September 23, 2021. - The human immune system has T lymphocytes to protect itself, which identify and destroy foreign cells. Cellectis develops CAR-T cells, or "chimeric antigen receptor T cells", a genetically engineered T cell for use in immunotherapy. The principle of the new and innovative class of therapy is to retrain the immune system to target a specific disease, such as cancer cells. Cellectis manufactures the RNA molecules in the Paris laboratory, which are then sent to a production site in the USA. After many more operations the RNA is turned into molecular scissors capable of cutting a fragment of DNA in T cells. (Photo by THOMAS COEX / AFP) (Photo by THOMAS COEX/AFP via Getty Images)
Image

Overview

  • C2S-Scale 27B, a 27‑billion‑parameter model built on Gemma, was designed to interpret the 'language' of individual cells.
  • The team used a dual‑context virtual screen across more than 4,000 compounds to pinpoint conditional amplifiers of immune activity.
  • The model predicted that the CK2 inhibitor silmitasertib (CX‑4945) would boost antigen presentation only under low‑interferon conditions.
  • Yale researchers validated the prediction in human neuroendocrine cell models not used in training, observing roughly a 50% increase in antigen presentation.
  • Google and Yale released a preprint and code for independent scrutiny, with researchers stressing that the findings are early and require peer review plus preclinical and clinical testing.