AbDiffuser: Full-Atom Generation of In Vitro Functioning Antibodies
Tuesday October 10th, 4-5pm EST | Karolis Martinkus, Machine Learning Scientist, Prescient Design, Genentech, Roche
Summary: We will discuss AbDiffuser, our latest equivariant and physics-informed diffusion model for the joint generation of antibody 3D structures and sequences. AbDiffuser is built on top of a new representation of protein structure, relies on a novel architecture for aligned proteins, and utilizes strong diffusion priors to improve the denoising process. Our approach improves protein diffusion by taking advantage of domain knowledge and physics-based constraints; handles sequence-length changes; and reduces memory complexity by an order of magnitude enabling backbone and side chain generation. We have validated AbDiffuser in silico and in vitro. Numerical experiments showcase the ability of AbDiffuser to generate antibodies that closely track the sequence and structural properties of a reference set. Laboratory experiments confirm that all 16 HER2 antibodies discovered using AbDiffuser were expressed at high levels and that 57.1% of selected designs were tight binders.
Karolis Martinkus is a Machine Learning Scientist at the Prescient Design team within Genentech Research & Early Development (gRED) where he works on generative models for de-novo antibody design. He completed his PhD at ETH Zurich under the supervision of Prof. Roger Wattenhofer, where he focused on applying deep learning to structured domains (e.g. graphs), in particular generative models.