Events

CRISPR genome editing has revolutionized life science research and has the potential to transform genomic medicine. In order to edit a genomic site, CRISPR-Cas proteins must first recognize a specific DNA motif called the PAM (Protospacer-Adjacent Motif). While researchers can program a guide RNA to target nearly any genomic site, PAM recognition is governed by protein-DNA interactions and limits the range of editable sites. At Profluent, we use machine learning to modulate protein-DNA interactions for diverse proteins. In this talk, I will describe Protein2PAM, a deep learning model that rapidly and accurately predicts PAM recognition directly from CRISPR-Cas proteins. Protein2PAM was trained on a dataset of over 45,000 CRISPR-Cas PAMs identified from systematic genome and metagenome mining. As a proof of concept for protein engineering, we employ Protein2PAM to computationally evolve Cas9 protein variants with broadened PAM recognition and enhanced cleavage rates. Our work represents the first successful application of machine learning to achieve customization of Cas enzymes for alternate PAM recognition, paving the way for personalized genome editing.

Register for the Livestream

About the Speaker

Stephen Nayfach leads Bioinformatics at Profluent Bio, where his team builds a platform for the curation of large-scale evolutionary and lab-generated data, enabling natural discovery and powering machine learning models. He holds a Ph.D. in Bioinformatics from UCSF, and was previously a Research Scientist at the Joint Genome Institute. During his career, he has contributed to pioneering research and developed impactful tools that have advanced microbial data science, metagenomics, and genome editing.