Rishi Bedi - Research

Protein-Protein Docking

As a research assistant in Ron Dror's group at Stanford, I'm working on deep learning approaches to predict the structure of protein complexes. More to come on this soon :)

Antibody Repertoire Analysis

In collaboration with Distributed Bio, I develop methods to enable the analysis of antibody repertoires.

I led the development of a method to allow the genome-free characterization of VDJ germline segments from high-throughput sequencing data of any species. If you're a biologist looking to characterize the immune system of an interesting species, please get in touch! I'm now exploring machine learning methods to characterize and explore antibody space.

Bifunctional Antibody-Ligand Traps (“Y-traps”)

In collaboration with a team at Johns Hopkins University, I developed novel antibody-ligand traps for the treatment of cancer. We started a company, Y-Trap, Inc. to advance these therapeutics into the clinic.

• Bifunctional immune checkpoint-targeted antibody-ligand traps that simultaneously disable TGFβ enhance the efficacy of cancer immunotherapy. Ravi R., Noonan K., Pham V., Bedi R., et al. Nature Communications, 2018.

• Inhibition of TGF-β enhances the in vivo antitumor efficacy of EGF receptor–targeted therapy. Bedi A., Chang X., Noonan K., Pham V., Bedi R., et al. Molecular Cancer Therapeutics, 2012.

ASSESS-MS

I spent a summer at Novartis in Basel, Switzerland working on the ASSESS-MS project, in collaboration with Microsoft Research London. ASSESS-MS is a tool that uses depth-sensing computer vision to automatically classify motor dysfunction in multiple sclerosis patients. Neurologists’ assessments of this motor dysfunction suffer from significant inter-rater (and even intra-rater) variability, so a computational approach could provide a more consistent and finer-grained evaluation of dysfunction.

My project focused on adapting TrueSkill, a Bayesian skill rating system developed by Microsoft, to tractably determinine ground truth labels using multiwise comparison data gathered from expert neurologists.

• Setwise comparison: Consistent, scalable, continuum labels for computer vision. Sarkar A., Morrison C., Dorn JF, Bedi R., et al. CHI, 2016.