Large-Scale Causal Gene Regulatory Network Inference from Genetic Interventions

Pooled genetic screens with single-cell transcriptomics readout promise to resolve the causal relationship between genes in a given population of cells. The most commonly employed statistical approaches to process this data, however, mainly consist of regression. Because of potential confounding factors, the results of such an analysis may not be interpreted as the (causally) direct effect of a gene onto other genes. Separately, the field of causal discovery learning recently introduced many algorithms that are applicable to inferring gene regulatory networks from Perturb-seq data, but most research has so far focused on relatively small causal graphs, with up to tens of nodes. Here, we introduce the notion of factor directed acyclic graphs (f-DAGs) as a way to restrict the search space to non-linear low-rank causal interaction models. Combining this novel structural assumption with recent advances that bridge the gap between causal discovery and continuous optimization, we propose Differentiable Causal Discovery of Factor Graphs (DCD-FG), a scalable implementation of f-DAG constrained causal discovery for high-dimensional interventional data. DCD-FG uses a Gaussian non-linear low-rank structural equation model and shows significant improvements compared to state-of-the-art methods in both simulations as well as a recent large-scale single-cell RNA sequencing data set with hundreds of genetic interventions.

Bio:
Romain Lopez is a postdoctoral fellow with a joint appointment between Genentech Research and Early Development and Stanford University, hosted by Aviv Regev & Jonathan Pritchard. He obtained a PhD degree in May 2021 from the department of Electrical Engineering and Computer Sciences at UC Berkeley, advised by Mike Jordan & Nir Yosef. His research interests lie at the intersection of statistics, computation and modeling with a strong focus on biological applications. Romain developed scVI and several extensions, a scalable set of core analysis tools for single-cell omics data, based on deep generative models. He is a lead contributor to scvi-tools, an open-source library for deep probabilistic modeling of single-cell data. He also worked on counterfactual inference and offline policy learning methods in collaboration with technology companies (Ant Financial, 2018 & Amazon, 2019). Before graduate school, Romain obtained a Diplome d'Ingenieur and a MSc in Applied Mathematics from Ecole polytechnique, Palaiseau.