Snorkel AI helps MSKCC streamline HER-2 patient identification

MSKCC needed to classify HER2 protein status in patient records to identify clinical-trial candidates, but manual review was slow since only physicians and researchers could label the complex unstructured data, creating bottlenecks in speed, consistency, and disagreement detection. Snorkel's team worked with MSKCC to build eight imperfect programmatic labeling functions that Snorkel combined to auto-label raw data, iteratively refined via model-guided error analysis that caught mislabeled ground truth, then trained an XGBoost model on the result. The final model reached 93% overall accuracy (92% on the test set) and an 87% average F1 score across classes, auto-labeling thousands of records in weeks instead of months; one engineer said a prior regex model took a long time to build, whereas with Snorkel "it was a few hours to get to the results we achieved."

Snorkel AI is a data-development and evaluation company whose programmatic labeling tools build training and benchmark datasets used to stress-test whether a model approach is viable before a full build.

This sits in the Design & Stress-Test stage of the AI use-case lifecycle: Named examples from the tools pharma teams use to scope, prototype and benchmark an AI use case before committing to build it — data-catalog, sandbox and evaluation platforms.