AMS Special Seminar Series | Surya Maddu

Location: Shaffer 3

When: January 29th at 1:30 p.m.

Title: Inference of Multiscale Dynamical Systems from Biological Data 

Abstract: Biological systems exhibit nonlinear, stochastic, and multiscale dynamics that challenge classical mathematical modeling approaches. Although modern experiments generate vast, high-dimensional datasets, the multimodal foundation models trained on them remain largely descriptive, lacking the ability to uncover causal dynamical mechanisms.

In this seminar, I will present our efforts to integrate biophysical modeling, statistical inference, and large-scale machine learning to bridge this gap. First, at the cellular scale, I will introduce Probability Flow Inference (PFI), a framework for learning biophysically consistent stochastic processes directly from time-resolved, cross-sectional omics data. I will present identifiability conditions under which the latent process is uniquely determined from static snapshots, and describe scalable, simulation-free inference strategies based on flow matching. Using cellular differentiation as a case study, I show how these models yield quantitative, testable predictions for in silico perturbations. Second, for spatially extended systems, I will discuss physics-informed neural networks (PINNs), highlighting theory-guided optimization strategies that mitigate stiffness in gradient flow dynamics, avoid catastrophic forgetting during sequential training, and improve convergence of Hamiltonian Monte Carlo (HMC) samplers, with applications from spatial biology. Finally, I will outline a theoretical framework for bridging scales and show how equivariant neural operators support the construction of accurate closure models that ensure consistency between microscopic physics and macroscopic dynamics.

Overall, this work establishes how theory and physics-guided statistical inference and machine learning provide a foundation for causal, predictive modeling of biological dynamics.

Bio: Surya is currently a Flatiron Research Fellow at the Center for Computational Biology (CCB) at the Flatiron Institute, Simons Foundation, and also holds an appointment as a Research Associate in the Department of Molecular and Cellular Biology at Harvard University as a CCBx Fellow. Previously, he was a Visiting Scholar at the NSF–Simons Center for Mathematical and Statistical Analysis of Biology in the Faculty of Arts and Sciences at Harvard University. He received his Ph.D. in Computer Science from the Max Planck Society and Technische Universität Dresden in 2021 and was also part of the International Max Planck Research School for Cell, Developmental and Systems Biology during his doctoral studies. He holds a Master’s degree in Computational Engineering and Applied Mathematics from Ruhr-Universität Bochum (2017) and a Bachelor’s degree in Mechanical Engineering from the National Institute of Technology, Karnataka (2014).

Zoom link: https://wse.zoom.us/j/92215066845