🔬 Training Transformers to solve 95% failure rate of Cancer Trials — Ron Alfa & Daniel Bear, Noetik

Cancer drug development frequently fails because clinical trials lack precise patient selection, relying on outdated cell line models that fail to translate to human biology. Noetik addresses this by building foundation models trained on proprietary, high-quality multimodal data, including H&E staining, spatial transcriptomics, and protein markers. By generating massive, spatially resolved datasets, these models simulate patient-level biology to identify distinct, therapeutically relevant cancer subtypes. This approach moves beyond simple biomarkers, allowing for the prediction of drug response across diverse patient populations. Utilizing autoregressive transformer architectures like Tario, the system scales across long-context tissue data to uncover complex, non-linear biological patterns. This methodology enables more effective trial design and therapeutic discovery, as evidenced by recent licensing agreements with major pharmaceutical partners like GSK, signaling a shift toward data-driven, patient-centric oncology research.