Sebastian Castillo-Hair, PhD
Postdoctoral researcher, University of Washington
Date: Friday, Jan. 16
Time: 9–9:50 a.m.
Location: SCOB 228
Faculty host: Xiaojun Tian
Abstract
Across tissues, developmental stages and disease conditions, cells adopt distinct states characterized by unique molecular profiles. The ability to write DNA- and RNA-encoded programs that sense and modulate cell states holds transformative potential for biotechnology, with applications that include gene therapies with cell type-specific action that minimize off-target effects, and guided stem cell differentiation for regenerative medicine. Yet, an incomplete understanding of how DNA and mRNA regulate core processes such as gene expression limits their rational design. Artificial Intelligence (AI) models that predict molecular function from sequence offer a powerful tool to overcome these barriers.
In this talk, I will present our recent work at the intersection of AI and high throughput experiments to study and engineer genetic programs that recognize cell types, tissues and developmental stages. First, I will describe our work on synthetic enhancers — DNA sequences that control transcription with cell type-specificity. Using models trained on public genomic data from hundreds of cell types, tissues and in vitro differentiated cells, we constructed an atlas of over 50,000 synthetic enhancers. We experimentally validated ~9,000 enhancers across ten human cell lines representing various tissues, as well as in mouse retinas, and found that they showed the expected specificity while greatly outperforming natural enhancer controls. Using explainable AI, we identified sequence features learned from natural enhancers and amplified by our models. This work resulted in the largest repository of AI-designed enhancers for cell type-targeting to date, and the most comprehensive experimental validation. Next, I will describe recent work on mRNA sequence engineering, where we leveraged AI models to optimize mRNAs for gene editing of tumor-targeting cell therapies, and high throughput experiments to derive rules for designing mRNAs with cell type-specific stability. Finally, I will describe results on modeling gene regulation during the complex cell state transitions that occur in embryonic development, and in using explainable AI to identify DNA and mRNA sequences that dynamically regulate gene expression. Our work shows the potential of combining AI, high throughput experiments and synthetic biology to decode and engineer cell state-responsive biological systems.
Biosketch
Sebastian Castillo-Hair is a Postdoctoral Researcher in Professor Georg Seelig’s group at the University of Washington. In his research, he uses machine learning and high throughput experiments to study gene regulation and design synthetic sequences with novel capabilities. He has applied this approach to optimize mRNA-delivered gene editing therapeutics, to engineer human enhancers with cell type-specific activity and to uncover the sequence determinants of differentiation and gene regulation during zebrafish development. Sebastian received his PhD in Bioengineering from Rice University, where he worked in Professor Jeffrey Tabor’s group on bacterial optogenetics and development of open-source tools for synthetic biology.
