Title: Sequence modeling from molecular to genome scale with Evo
Speaker: Brian Hie, PhD
Affiliation: Stanford University, Chemical Engineering and Data Science
Position: Assistant Professor
Host: Jiening Zhu, PhD – Gerber Lab
Date: Monday November 18, 2024
Time: 4:00-5:00PM ET
Zoom: https://partners.zoom.us/j/82163676866
Meeting ID: 821 6367 6866
Abstract: The genome is a sequence that encodes the DNA, RNA, and proteins orchestrating an organism’s function. We present Evo, a long-context genomic foundation model with a frontier architecture trained on millions of prokaryotic and phage genomes, and report the first scaling laws on DNA to complement observations in language and vision. Evo generalizes across DNA, RNA, and proteins, enabling zero-shot function prediction competitive with domain-specific language models and the generation of functional CRISPR-Cas and transposon systems, representing the first examples of protein-RNA and protein-DNA co-design with a language model. Evo also learns how small mutations affect whole-organism fitness and generates megabase-scale sequences with plausible genomic architecture. These prediction and generation capabilities span molecular to genome scales of complexity, advancing our understanding and control of biology.
Brian Hie is an Assistant Professor of Chemical Engineering at Stanford University, the Dieter Schwarz Foundation Stanford Data Science Faculty Fellow, and an Innovation Investigator at Arc Institute, where his group conducts research at the intersection of biology and machine learning.
Fibrosis is a pathology of excessive scarring which causes morbidity and mortality worldwide. Fibrosis is a complex process involving thousands of factors, therefore, to better understand fibrosis and develop new therapeutic approaches, it is necessary to simplify and clarify the underlying concepts. In this talk, I will introduce a mathematical model we recently developed for a cell circuit between myofibroblasts and macrophages – the two cell types that produce and remodel the scar. The mathematical framework predicts two types of fibrosis – hot fibrosis with abundant macrophages and myofibroblasts, and cold fibrosis dominated by myofibroblasts alone. Moreover, we use the model to predict that the autocrine signal for myofibroblast division is a potential therapeutic target to reduce fibrosis. Finally, I will discuss how we use myocardial infarction (MI), a widely studied in-vivo injury model for cardiac fibrosis, to test these theoretical concepts and intervention strategies experimentally.
Spatial biology is a new frontier that has become accessible through advances in spatial profiling technologies, such as multiplexed in situ imaging spatial proteomics, which can provide single-cell resolution up to 60 markers. In this talk, I will introduce a computational analysis pipeline that performs integrative analysis of spatial proteomics and single-cell RNA sequencing to identify clinically-relevant cellular interactions. The pipeline features (1) CELESTA, an unsupervised machine learning method for cell type identification in multiplexed spatial proteomics data; (2) a geospatial statistical method to identify cell-cell colocalizations; and (3) an integrative coupling of spatial proteomics and single-cell RNA sequencing data that identified cell-cell crosstalk associated with lymph node metastasis in head and neck cancer which we have validated through mouse model studies.
Microbiome studies hold tremendous potential along with substantial computational challenges. I will present two computational approaches for microbiome data analysis. First, I will present SCRuB, a new method for in silico removal of contamination from microbiome data. We show that modeling the taxonomic composition of contamination sources, rather than trying to infer whether specific taxa are categorically contaminant, allows for more robust decontamination and improves downstream phenotypic prediction. Second, I will present copangraph, a new graph-based approach for representing genomic variability within and across microbiomes. We demonstrate a hybrid co-assembly approach that yields high-quality representation of the microbiome.