Extracting causal gene perturbations from transcriptomic data

Cell fate transitions hinge on the relationship between genotype and phenotype. This relationship remains an outstanding question for many traits and transitions between them because such traits are generally complex, especially in the case of transitions to and from disease states. The challenge arises from complex traits being determined by a combination of multiple genes (or loci), which leads to an explosion of possible genotype-phenotype mappings. Here, we develop two hypothesis-test-free approaches that leverage transcriptional perturbation data endowed with causal information and generative machine learning models. Our implementation of the first approach includes a variational autoencoder trained on human transcriptional data, which is incorporated into an optimization framework. The approach generates trait expression profiles from which we conduct constrained optimization to find the gene perturbations whose measured transcriptomic responses best explain trait differences. By considering several complex disease traits, we show that the approach identifies causal genes that cannot be detected by the primary existing techniques. In our second approach, which we call ExPert, we directly classify transcriptional perturbations from gene expression profiles in order to infer which causal perturbations cause exotic transcriptional fate transitions. We find that we can accurately classify both gene expression and ATACseq profiles to uncover hidden perturbations.