About
We build generative models of biological systems, trained via reinforcement learning on indirect observation data. This approach enables flexibility, interpretability, and biological grounding.
Flexibility
GReinSS offers the flexibility of modern machine learning to match nearly any biological generative process. In contrast to single-pass prediction methods, generative reinforcement learning can model any sequential process of generating complex combinatorial states with biological structure.
Interpretability
Existing machine learning methods learn latent states in the form of arbitrary continuous representations of the data rather than directly interpretable biological states. Instead, GReinSS directly defines latent states as the real unobserved biological states of interest. This makes the predictions of GReinSS directly interpretable and applicable biological outputs.
Biological Grounding
The process of generating states and observations is directly based on the biology of the particular system being modeled. This gives the opportunity to apply domain knowledge to biologically ground these processes, instead of treating each dataset as an unstructured learning problem separate from known biology.
A message from our CEO
Stefan Ivanovic, Ph.D. Founder & CEO, ReinBio
Research expertise in reinforcement learning, self-supervised learning, genomics, and cancer biology. Additional background in algorithm development, theoretical physics, and pure mathematics.
“At Rein Bio, our mission is to help solve the most important problems in computational biology by providing biologically grounded modern machine learning. Our approach has enabled biological discovery across a wide range of multi-omics applications, including tumor phylogenetics, single-cell genomics, and transcriptomics. We believe the future of computational biology lies in combining modern machine learning with models grounded in real biological structure.”