Bioinformatics Seminar

All of life encodes information with DNA. While tools for sequencing, synthesis, and editing of genomic code have transformed biological research, intelligently composing new biological systems would also require a deep understanding of the immense complexity encoded by genomes. We introduce Evo 2, a biological foundation model trained on 9.3 trillion DNA base pairs from a highly curated genomic atlas spanning all domains of life. We train Evo 2 with 7B and 40B parameters to have an unprecedented 1 million token context window with single-nucleotide resolution. Evo 2 learns from DNA sequence alone to accurately predict the functional impacts of genetic variation—from noncoding pathogenic mutations to clinically significant BRCA1 variants—without task-specific finetuning. Applying mechanistic interpretability analyses, we reveal that Evo 2 autonomously learns a breadth of biological features, including exon–intron boundaries, transcription factor binding sites, protein structural elements, and prophage genomic regions. Beyond its predictive capabilities, Evo 2 generates mitochondrial, prokaryotic, and eukaryotic sequences at genome scale with greater naturalness and coherence than previous methods. Guiding Evo 2 via inference-time search enables controllable generation of epigenomic structure, for which we demonstrate the first inference-time scaling results in biology. We make Evo 2 fully open, including model parameters, training code, inference code, and the OpenGenome2 dataset, to accelerate the exploration and design of biological complexity.

Recent years have seen dramatic advances in both experimental determination and computational prediction of macromolecular structures. These structures hold great promise for the discovery of highly effective drugs with minimal side effects, but structure-based design of such drugs remains challenging. I will describe recent progress toward this goal, using both atomic-level molecular simulations and machine learning on three-dimensional structures.

Biological replicators are locked in deeply intertwined genetic conflicts with each other. Using comparative genomics, protein sequence and structure analysis and evolutionary investigations, my lab has uncovered a staggering diversity of molecular armaments and mechanisms regulating their deployment, collectively termed biological conflict systems. These include toxins used in interorganismal interactions and a host of mechanisms involved in self/nonself discrimination, especially in the context of host-selfish element conflicts. Our studies have helped identify shared syntactical features in the organizational logic of biological conflict systems. These principles can be exploited to discover new conflict systems through computational analyses. Further, we find that across the range of biological organization, from intragenomic conflicts to interorganismal conflicts, a circumscribed set of effector protein domain families is deployed, targeting genetic information flow through the Central Dogma, certain membranes, and key molecules like NAD+ and NTPs. This has led to significant advances in discovering new biochemistry of these systems and furnished new biotechnological reagents for genome editing, sequencing and beyond. I’ll discuss this using specific examples of toxins in interorganismal conflict and effectors in antiviral immunity.

Human genomics has relied on a single reference genome for the last twenty years. This reference genome is a corner stone of much of what we do in genomics but it can not, by definition, represent the variation present in the human population, and as a reference introduces a pervasive bias into genomic analyses. I will survey our recent efforts, through the Human Pangenome Reference Consortium, to build and use a reference pangenome – a collection of extremely high-quality reference genomes related together by a consensus genome alignment that we intend as a replacement for the reference genome.