UC Berkeley

Understanding how gene regulatory systems evolve and how they can be engineered requires integrating scalable molecular technologies with predictive algorithms capable of operating across evolutionary space. This dissertation develops high-throughput experimental and computational frameworks to elucidate and reprogram gene regulation across bacteria, fungi, and plants. Directed evolution of plasmid origins of replication demonstrated that tuning copy number enhances genetic transformation efficiency across fungal and plant hosts. Genome-wide transcription start site mapping using differential RNA sequencing and an optimized identification model resolved cis-regulatory architecture at nucleotide resolution across the three major agrobacterial lineages, revealing lineage-specific promoter structure and divergence in cell cycle and virulence circuits, while highlighting the importance of algorithmic parameter optimization for accurate genome-level regulatory inference. To extend discovery of evolutionary-level gene regulation, an active learning framework integrating protein language models with a deep ensemble of residual neural networks quantitatively predict trans-regulatory strength and guided uncertainty-driven sampling across the fungal branch of life, expanding functional annotations and uncovering underrepresented biochemical codes underlying transcriptional activation. To interrogate and engineer transcriptional regulators at scale, ENTRAP-seq was developed to multiplex measurement of regulatory activity from thousands of protein variants directly in planta, enabling discovery of previously unannotated activation domains and machine-guided design of transcription factor activity. Together, these studies establish a generalizable strategy in which molecular measurement and algorithmic modeling operate in a closed loop, enabling systematic discovery of evolutionary regulatory principles and rational engineering of gene expression programs across kingdoms of life.