Geometric Deep Learning for Antibody Drug Discovery Wengong Jin Postdoctoral Fellow, Eric and Wendy Schmidt Center at Broad Institute Modeling antibody-antigen binding is pivotal to drug discovery. Geometric deep learning is a promising paradigm for binding energy prediction, but its accuracy is limited by the size of training data, as high-throughput binding assays are expensive. Herein, we propose an unsupervised binding energy prediction framework named DSMBind, which does not need experimental binding data for training. DSMBind is an energy-based model that estimates the likelihood of a protein complex via SE(3) denoising score matching (DSM). This objective, applied at both backbone and side-chain levels, builds on a novel equivariant rotation prediction network derived from Euler's Rotation Equations. We find that the learned log-likelihood of protein complexes is highly correlated with experimental binding energy across multiple antibody-antigen binding prediction benchmarks. We further demonstrate DSMBind's zero-shot binder design capability through a PD-L1 nanobody design task, where we randomize all three complementarity-determining regions (CDRs) and select the best CDR sequences based on DSMBind score. We experimentally tested the designed nanobodies with ELISA binding assay and successfully discovered a novel PD-L1 binder. In summary, DSMBind offers a versatile framework for binding energy prediction and binder design.
Unbiased Biological Discovery Without a Reference Genome Tavor Baharav Postdoctoral Fellow, Eric and Wendy Schmidt Center at Broad Institute Computational genomics pipelines often rely heavily on alignment of sequencing data to a reference genome; however, this use of a reference genome can bias downstream inference and limit discovery of novel biology. In this talk I will discuss a unifying paradigm for genomic inference, SPLASH, which performs inference directly on raw sequencing data. We demonstrate SPLASH’s power for unbiased discovery by identifying viral strain mutations, cell-type-specific isoforms, and Ig and TCR diversity, in addition to tissue-specific transcripts in octopus and geographic and seasonal variation and diatom association in eelgrass.
Optimal Experimental Design for Genetic Perturbations Jiaqi Zhang Ph.D. Candidate (PI Caroline Uhler), MIT Department of Electrical Engineering and Computer Science (EECS) and The Eric and Wendy Schmidt Center at Broad Institute Sequential experimental design to discover interventions that achieve a desired outcome is a key problem in various domains. A predominant example is how to identify optimal genetic perturbations that induce a specific cell state transition. This talk covers our methods for predicting unseen combinatorial perturbational effects and actively selecting the next most-informative experiment for identifying desirable interventions more efficiently.