Escape Labs

Escape Labs is an autonomous cancer-research workflow built using published FGFR laboratory data. Its ambition is to help researchers anticipate drug resistance and prioritize the next experiment. Nemotron on Crusoe proposes bounded modeling experiments, numerical tools train predictors, and an independent evaluator measures results, accepts or rejects revisions, and preserves the evidence. A separate improvement to our FGFR2 sequence classifier raised mean-fold average precision from 0.09732 to 0.11135, a 14.4% relative gain. It identified four highly activating mutations instead of one within the same 68-candidate budget, measured across 668 variants and three fixed development folds. We also tested GLM-5.3 as an alternative planner and integrated reviewed research evidence from You.com. We built the data pipeline, experiment vault, agent loop, evaluation tools, and demo during the hackathon, using published assays and pretrained models. The public repository records all experiments and results. Preclinical research only; no patient inputs or treatment recommendations.