Learning causal networks with latent variables from multivariate information in genomic data

Learning causal networks from large-scale genomic data remains challenging in absence of time series or controlled perturbation experiments. We report an information- theoretic method which learns a large class of causal or non-causal graphical models from purely observational data, while including the effects of unobserved latent variables, commonly found in many genomic datasets. Starting from a complete graph, the method iteratively removes dispensable edges, by uncovering significant information contributions from indirect paths, and assesses edge-specific confidences from randomization of available data. The remaining edges are then oriented based on the signature of causality in observational data. The approach and associated algorithm, miic, outperform earlier methods on a broad range of benchmark networks. Causal network reconstructions are presented at different biological size and time scales, from gene regulation in single cells to whole genome duplication in tumor development as well as long term evolution of vertebrates. Miic is publicly available at https://github.com/miicTeam/MIIC.Author summary: The reconstruction of causal networks from genomic data is an important but challenging problem. Predicting key regulatory interactions or genomic alterations at the origin of human diseases can guide experimental investigation and ultimately inspire innovative therapy. However, causal relationships are difficult to establish without the possibility to directly perturb the organisms’ genome for ethical or practical reasons. Besides, unmeasured (latent) variables may be hidden in many genomic datasets and lead to spurious causal relationships between observed variables. We propose in this paper an efficient computational approach, miic, that overcomes these limitations and learns causal networks from non-perturbative (observational) data in the presence of latent variables. In addition, we assess the confidence of each predicted interaction and demonstrate the enhanced robustness and accuracy of miic compared to alternative existing methods. This approach can be applied on a wide range of datasets and provide new biological insights on regulatory networks from single cell expression data or genomic alterations during tumor development. Miic is implemented in an R package freely available to the scientific community under a General Public License.