GSK

Multimodal foundation model for cardiac biosignals, pretrained by masked modeling on ECG, PPG, and clinical text from ~1.7 million individuals.

Phenotypic screening model for neuronal activity, encoding network-level calcium dynamics of iPSC-derived neurons into single-cell embeddings.

Genetic language model predicting disease risk and cell-type-specific expression changes from up to 88 megabases of an individual's genome sequence.

Antibody language model trained on paired and unpaired OAS sequences to suggest non-germline mutations instead of reverting them to germline.