TITLE

AI for Cellular Dynamics: From Generative Models to Precision Therapies

PRESENTER

Prof. Elham Azizi, Columbia University, USA

DESCRIPTION

Foundations of AI for Cellular Dynamics
We will begin with statistical and probabilistic foundations for modeling high-dimensional biomedical data. Topics include probabilistic graphical models, Bayesian inference, and optimal transport theory for comparing cellular distributions. I will introduce how these frameworks provide the mathematical basis for decoding cellular dynamics and interactions using single-cell and spatial omics, particularly in complex tissues like tumors.

Deep Generative Models for Multi-Modal Integration
This session will focus on deep learning-based generative models such as VAEs and multi-view architectures, designed for integrating diverse data types including single-cell RNA, ATAC, and spatial transcriptomics. I will highlight case studies where these models capture metabolic and immunosuppressive hubs in breast cancer and reconstruct cell-state trajectories in leukemia. The session will also cover representation learning strategies for aligning omics with histology imaging, drawing on emerging ideas in foundation models.

Transformers and Attention Mechanism for Cell-Cell Communication
Here I will present probabilistic and attention-based architectures, including transformers and graph neural networks, for modeling cellular interactions in tissue microenvironments. We will see applications in identifying T-cell subsets critical for leukemia therapy response and uncovering communication hubs that drive resistance. The second part will introduce causal discovery frameworks and Bayesian tools that help disentangle genetic versus environmental contributions to phenotypic plasticity, with examples from melanoma immunotherapy.

AI for Spatio-Temporal Dynamics and Precision Therapies The final session will bring together temporal modeling and cutting-edge AI. We will explore optimal transport for trajectory inference, hidden Markov models, Gaussian processes, and diffusion/flow-matching models for spatio-temporal single-cell data. Case studies will show how computational tools reconstruct gene regulatory networks involved in leukemia initiation and predict therapy responses across cancers. The lecture concludes with a forward look at how AI foundation models, tailored to biological contexts, can simulate tumor evolution and guide precision therapies.

ABOUT THE SPEAKER

Elham Azizi is the Herbert and Florence Irving Associate Professor of Cancer Data Research (in the Irving Institute for Cancer Dynamics) and Associate Professor of Biomedical Engineering at Columbia University. She is also affiliated with the Department of Computer Science, Data Science Institute, and the Herbert Irving Comprehensive Cancer Center. Elham holds a BSc in Electrical Engineering from Sharif University of Technology, an MSc in Electrical Engineering and a PhD in Bioinformatics from Boston University. She was a postdoctoral fellow in the Dana Pe'er Lab at Columbia University and Memorial Sloan Kettering Cancer Center. Her multidisciplinary research utilizes novel machine learning techniques and single-cell genomic and imaging technologies to study the dynamics and circuitry of interacting cells in the tumor microenvironment. She is a recipient of the Vilcek Prize for Creative Promise in Biomedical Science, Takeda/NYAS Early-Career Innovator in Science Award, Allen Distinguished Investigator Award, Chan Zuckerberg Initiative SDL Award, NSF CAREER Award, Tri-Institutional Breakout Prize for Junior Investigators, NIH NCI Pathway to Independence Award, American Cancer Society Postdoctoral Fellowship, and IBM Best Paper Award at the New England Statistics Symposium.