Designing Programmable Biologics with Generative Sequence Models
In this talk, I will share how my lab develops discrete generative models to design functional biologics for disease and bioremediation. Our work has centered on language models that de novo design peptides to bind and modulate undruggable targets, with experimental validation across rare neurodegenerative disorders, pediatric cancers, and viral infections. Because therapeutic design depends on clinically-viable properties beyond binding (solubility, half-life, non-toxicity), we have developed discrete diffusion algorithms to generate peptides, proteins, mRNAs, and heavy metal sequestrants that are Pareto-optimal across these properties. We have extended these frameworks to discrete flow matching models that generate isoform-specific, domain- and motif-resolved binders under competing therapeutic objectives, enabling the design of potent inhibitors and CAR T cell ligands. Finally, we have recently pioneered Schrödinger Bridge Matching, a new class of generative models that capture both biological states and the trajectories connecting them, from protein folding to drug-induced cell-state transitions, establishing a unified, programmable framework for molecular modeling and design.
Please use this link to access Zoom.
For any questions, please contact Hyacinth Camillieri at hcamillieri@gc.cuny.edu