Brian Hie at Stanford discusses training Evo2, a genomic foundation model, with 9.3 trillion nucleotides from over 128,000 genomes.

Protein sequence design using artificial intelligence (e.g. AlphaFold) has advanced significantly in recent years. However, there are still major challenges for nucleic acid sequence design by artificial intelligence. This is mainly due to lack of nucleic acid regulatory element understanding as these sequences fall within the Cis-Regulatory elements, out of the open reading frame. To address these challenges, researchers at Agricultural Genomics Institute of Shenzhen have developed CisEncoder, a platform that combines high-throughput experimental data with deep learning models to understand the sequence-to-activity relationship of the regulatory elements. With clonal genes by Twist Bioscience, the researchers have validated this new platform across agricultural biology models and animal models.

This application note demonstrates how Twist’s precisely synthesized Site-Saturation Variant Libraries (SSVLs) offer a scalable and precise platform to dissect complex mutation interactions and their functional consequences. Using the SARS-CoV-2 spike protein’s receptor binding domain (RBD) as a case study, the Starr Lab at the University of Utah applied SSVLs to rapidly generate and test every possible amino acid substitution across multiple variant backgrounds.