Hematology

Hematological malignancies present a paradox: molecularly rich data environments with persistently high clinical trial failure. The biology is well-characterized in cell lines, but lacking translation to clinical outcomes. What's missing is a system that models therapeutic response across the full molecular context of individual patients.


We train virtual patient models on matched longitudinal datasets from patients with various hematological malignancies — integrating clinical history, mutations, methylation, gene expression, and clinical endpoints to predict which patients will respond and why.

Across hematological malignancies, our models recover prognostic subgroups learned entirely from multimodal data, while also stratifying risk better than clinical standards and genomic features alone.


A virtual patient model trained on paired-multimodal profiles for multiple downstream use cases including patient selection, reverse translation, predicting time to progression event, and overall response to standard-of-care regimens with clinically meaningful accuracy.