Abstract
Modeling the dynamics of complex real-world systems from temporal snapshot data is crucial for understanding phenomena such as gene regulation, climate change, and financial market fluctuations. Existing methods based on the Schrödinger Bridge or Flow Matching are limited in their ability to combine data from multiple time points and experimental settings. We propose Multi-Marginal Flow Matching (MMFM), which constructs a flow using smooth spline-based interpolation across time points and conditions, and regresses it via classifier-free guided Flow Matching. This framework shares contextual information across trajectories and significantly outperforms existing methods at imputing data at missing time points, including on a single-cell genomics dataset with chemical perturbations across time.
Romain Lopez
Assistant Professor of Computer Science and Biology
My research interests lie at the intersection of statistics, computation and modeling of biological data. A significant part of my research is driven by building more statistically accurate and faster machine learning software for analyzing genomics data.