ASSET Seminar: “Discrete Generative Models for Programmable Molecule Design”

Abstract: The Chatterjee Lab at the University of Pennsylvania develops discrete generative models that design functional biologics directly from sequence. Our work spans a new class of algorithms that unify discrete diffusion, flow matching, and Schrödinger bridge formulations for controllable, multi-objective molecular generation. Beginning with masked discrete diffusion frameworks such as PepTune, SOAPIA, NOSIE, Metalorian, and MemDLM, we demonstrated that discrete denoising processes can generate peptides, nanobodies, and membrane proteins with experimentally validated functionality. We then extended this paradigm through flow-based optimization, where models like Gumbel-Softmax Flow, MOG-DFM, and AReUReDi perform Pareto-efficient multi-objective design of therapeutic molecules across competing biochemical properties such as binding, solubility, and stability. Finally, our recent Schrödinger-bridge frameworks (BranchSBM, EntangledSBM, and ScooBDoob) formulate molecular and cellular dynamics as optimal transport between distributions, enabling the simulation, control, and design of protein folding and cell-state transitions. Together, these advances establish a unified sequence-based foundation for programmable biologic design, linking diffusion, flow, and bridge theories to the creation of next-generation therapeutic peptides and proteins.

Pranam Chatterjee is an Assistant Professor of Bioengineering and Computer and Information Science at the University of Pennsylvania. His laboratory, the Programmable Biology Group, develops new methods for de novo protein and peptide design, integrating new algorithmic theory for generative sequence modeling with translational in vitro and in vivo experimental platforms. Specifically, the lab uses these tools for diverse protein modulation, genome editing, and cell engineering applications, with particular focus on developing therapeutics for rare pediatric neurodegenerative diseases and cancers and molecules for bioremediation. He received his SB, SM, and PhD from MIT, and their work has been awarded multiple NIH and foundation grants. He is also very passionate about translating his academic research to the clinic as efficiently as possible. As such, he co-founded three companies, Gameto Inc., UbiquiTx Inc., and AtomBioworks Inc., which are building upon his foundational research to develop novel protein-based cancer therapeutics and fertility solutions.