Physical cartography of the cell: Immune-cancer perturbation maps from statistical-physics and machine learning
Drug resistance, immune exhaustion, and metastasis all arise from rare cells whose gene regulatory networks are driven into clinically deleterious states by genetic and chemical perturbations. My research program aims to identify causal drivers of these transitions by developing interpretable, physics-based AI and optimization frameworks that transform existing single-cell data into scalable in silico functional screens in three steps. 1) Recasting condition-to-condition transcriptomic differences as an explicit optimization problem in gene expression space allows me to identify perturbations that combine to most effectively steer cellular states from one phenotype to another, such as from drug-sensitive to resistant. 2) To complement this approach, I develop generative classification and regression methods that provide calibrated likelihoods and transparent attribution of predictions to underlying genes, regulatory programs, and other molecular features. 3) I integrate these components into a unified, cell-type–agnostic latent space learned from large public Perturb-seq atlases, enabling perturbations to be compared, ranked, and composed across diverse biological contexts without new experiments. Together, my work establishes a general paradigm for uncovering causal regulatory drivers of cell state transitions—drivers that are likely hiding in plain sight within the vast landscape of existing functional perturbation screens.
Hybrid, but in-person attendance is encouraged
https://bit.ly/BenKuznets-Speck