Associate Professor of Cancer Data Research and Biomedical Engineering at Columbia University

Machine Learning for Decoding Tumor Microenvironment Dynamics

The tumor microenvironment (TME) profoundly shapes tumor progression and response to therapies. My lab develops rigorous machine learning methods to unravel complex tumor-immune interactions using single-cell and spatial multi-omics data from patient specimens. In this talk, I will discuss computational frameworks designed to decode spatial and temporal dynamics of the TME, integrating spatial transcriptomics, histology imaging, and single-cell sequencing. I will present deep generative models for representation learning and multi-modal integration, enabling us to identify key metabolic and immunosuppressive hubs associated with aggressive subtypes of breast cancer. I will also showcase probabilistic and attention-based models that pinpoint crucial cell-cell interactions and T cell subsets critical for immunotherapy response in leukemia. Moreover, I will introduce causal discovery approaches and Bayesian tools to disentangle genetic and environmental drivers of phenotypic plasticity linked to immunotherapy resistance in melanoma. Finally, our computational tools reconstruct cell-state trajectories, uncovering dysregulated gene regulatory networks involved in leukemia initiation. Altogether, our machine learning toolkit has the potential to advance precision cancer therapies through decoding the principles of TME dynamics.