Mapping Genetic Interactions at Scale with Active Perturb-seq

Speaker

Prof. Romain Lopez

Assistant Professor
Department of Computer Science and Biology
NYU Courant Institute of Mathematical Sciences

Abstract

Genetic interactions (GIs) encode the combinatorial logic by which genes jointly control cellular state, but their systematic mapping remains experimentally intractable. Exhaustively assaying all ~200 million pairwise combinations among the ~20k human protein-coding genes is impossible, and high-dimensional molecular readouts have so far been restricted to small numbers of gene combinations. Here, we develop a lab-in-the-loop active learning framework for combinatorial Perturb-seq that iteratively couples predictive modeling with experimental design, enabling principled exploration of large GI spaces from sparse measurements. Applying this framework to a 600 × 600 gene-gene combinatorial space in K562 cells, we generate the largest transcriptome-scale GI map in human cells to date while assaying ~1.5% of 180,000 total possible pairs. A diversity-based active learning sampling technique consistently outperformed random sampling in recovering unseen GI structure, demonstrating that the transcriptional GI landscape is structured and learnable. Experimentally, active learning enriched for rare, high-signal interactions and accessed novel epistatic subspaces—revealing emergent epistasis gene programs not identifiable from the initial seeding round. By coupling predictive modeling with iterative experimentation, this framework makes large-scale, high-dimensional GI mapping tractable and provides a general paradigm for adaptive discovery across combinatorial biological spaces.

Romain Lopez is an Assistant Professor of Computer Science and Biology at New York University. He received his MSc in Applied Mathematics from École polytechnique and his PhD in Computer Science from UC Berkeley, advised by Professors Michael I. Jordan and Nir Yosef. He was a Postdoctoral Fellow at Genentech and Stanford University, hosted by Professors Aviv Regev and Jonathan Pritchard. He has received Best Paper honors at leading AI and computational biology venues, an AI2050 Early Career Fellowship, as well as a STAT Wunderkind Award, recognizing North America’s most promising early-career life scientists. His research develops probabilistic and causal machine learning methods to uncover the mechanisms that govern cellular behavior and disease, including the scVI framework and the scvi-tools ecosystem for deep generative modeling of single-cell omics data.