Probing the biomolecular design space with generative modeling

Discovering new-to-nature biomolecules requires navigating a design space that has only been sparsely explored. Traditional paradigms such as rational design, directed evolution, and high-throughput screening have enabled the development of many successful functional biomolecules, but are significantly limited when given an unseen target or novel function, where there is no starting point for design. Generative modeling offers a way to approximate the continuous manifold underlying observed biological data, enabling the sampling of new candidates beyond any finite dataset while remaining within the support of viable, functional biomolecules. Here, we outline recent progress in generative biomolecular design through an algorithmic lens, bridging theoretical foundations with its extensions and applications in truly de novo and controllable biomolecular design.