C6 - Gene Targeting and Gene Correction New Technologies
13: Language Models Generate Novel Genome Editors from Scratch
Type: Oral Abstract Session
Presentation Details
Session Title: New Technologies for Gene Targeting and Gene Correction
Gene editing has the potential to cure disease yet requires complex molecular machinery for precise manipulation of cellular genetic material. Editors derived from microbial CRISPR systems, while powerful, often face tradeoffs in key attributes like size, activity, and immunogenicity when used in non-native environments. AI-based protein engineering is a powerful alternative that has the potential to bypass evolutionary constraints and generate proteins with optimal properties. Here, we present the first programmable gene editor designed 'from scratch' using artificial intelligence models. Our models were trained on massive-scale biological diversity, including a dataset of over one million CRISPR operons systematically identified from 26 terabases of assembled genomes and meta-genomes. Our models generate thousands of diverse CRISPR-associated proteins never before seen in nature, including highly functional Type II nucleases that are >400 mutations away from any known natural protein. We also used AI models to generate fully synthetic single-guide RNAs that were compatible with their cognate gene editors and boosted activity. Our most performant AI-generated editor, which we denote OpenCas, exhibited up to 95% editing efficiency across a variety of cell types, has a low off-target rate, is compatible with base editing, and is predicted to have reduced immunogenicity. We intend to release OpenCas to the public in order to democratize the ethical usage of gene editing across research and commercial applications. Collectively, our results demonstrate the power of AI to build customized gene editing proteins and will be a foundation for AI-based protein engineering in the future.
Plain Language Summary
The team at Profluent Bio designed completely new gene modifiers using the latest advances in Artificial Intelligence and are releasing the best performing version called ""OpenCas"" to the public to democratize gene modifying technologies and find cures for genetic diseases faster.
Jeffrey A. Ruffolo, Stephen Nayfach, Aadyot Bhatnagar, Joseph Gallagher, Joel Beazer, Jennifer Yip, Riffat Hussain, Jordan Russ, Alexander J. Meeske, Peter Cameron, Ali Madani
Profluent Bio, Berkeley, CA"
Find This Session