PREDICTS puts 62 molecular models into a four-organization pilot

The September 24 pilot announcement creates a setting for comparing predictions with new experimental measurements; it does not yet report that validation.

The models address ADMET: absorption, distribution, metabolism, excretion and toxicity. These properties help determine whether a promising molecule can become a useful medicine. The named participants are ImmVue Therapeutics, Synko Biotherapeutics, Sanford Burnham Prebys and SyzOnc. According to Deep Origin, their work is intended to feed experimental observations back into model evaluation and development.

Availability and validation must remain separate. Sixty-two models means that a collection has reached partners; it does not tell us how accurately any one model predicts an unfamiliar compound, how uncertainty is calibrated or whether its use improves a development decision. The release is a company account, and the reported deployment is not independent prospective evidence of clinical benefit.

The wider setting helps explain the ambition. ARPA-H’s CATALYST programme aims to develop and validate human-relevant models for evaluating candidate medicines. Its scope spans data-driven prediction, living experimental systems and computational representations of human physiology. That is a research programme toward better evidence, not a declaration that a digital model can already replace the relevant laboratory or clinical assessment.

For a small biotechnology team, the accessible next step is specific: ask which endpoint is available, what inputs it accepts and what evaluation can be run on held-out compounds. Deep Origin advertises API access by request. That access route should not be confused with an open release of every PREDICTS model or its training data.