Biology Foundation Models

Bioptimus (Paris, spun off from Owkin) is creating foundation models trained not on internet text but on biological data: DNA sequences, protein structures, cell microscopy images, and clinical datasets. The goal is a general-purpose biological AI that can predict interactions across the entire molecular landscape of human disease.

The approach treats biology as a language problem: just as GPT learned language patterns from text, Bioptimus learns biological patterns from molecular and cellular data. Access to the Jean Zay supercomputer (France's national AI compute resource) enables training at the scale required for biological foundation models — a public-private partnership model distinctive to France.

Bioptimus sits in a Paris AI-for-biology cluster that includes Owkin (its parent company), Inria's computational biology teams, and Institut Pasteur. The combination of world-class biology research institutions and frontier AI talent makes France a natural home for the convergence of AI and life sciences. If successful, biological foundation models could compress drug discovery timelines from decades to years by predicting which molecules will work before synthesis.