BioML Seminar 5.1 - Codesigning protein sequence and structure with test-time search

​Welcome back to the seminar series from the BioML group in Machine Learning at Berkeley, sponsored by Amplify Partners. This week, we're excited to host Danny Reidenbach, a Senior Research Scientist at NVIDIA!

​Abstract: Today's most powerful AI systems pair generative models with adaptive test-time scaling or reasoning. This talk presents Proteína-Complexa, a model that codesigns sequence and structure in a continuous latent space with reward-guided search at inference time. I'll present wet-lab validation across a massive-scale benchmark of over one million designs against 127 targets, plus binders to PDGFR, ActRIIA, kinase targets, a viral glycoprotein, and a free carbohydrate — the first de novo proteins shown to bind a non-protein, polar target class.

​Speaker Bio: Danny Reidenbach is a Senior Research Scientist at NVIDIA, where he develops open models for atomistic and biological design. He obtained his M.S. in EECS from UC Berkeley, where his thesis focused on structure-based generative modeling techniques for molecular design, and holds a B.S. in Chemical Engineering and a B.A. in Computer Science, also from UC Berkeley.