Generative Modeling of Single Cell Gene Expression for Dose-Dependent Chemical Perturbations

Abstract

Single cell sequencing provides a new opportunity to study the heterogeneity of chemical perturbation within tissues. However, exploring the combinatorial space of all cell type-chemical combinations is experimentally and financially unfeasible. This space is significantly expanded by the dose axis of chemical perturbation. Thus, computational tools are needed to predict responses not only across tissues, but also across doses while capturing the nuances of cell type specific gene expression. Variational autoencoders simplify the single cell expression space allowing cross cell type predictions using simple vector arithmetic. However, differing sensitivities and non-linearities make cell type specific gene expression predictions following treatment at higher doses challenging. Here we introduce single cell Variational Inference of Dose-Response (scVIDR) which achieves high dose and cell type specific predictions better than other state of the art algorithms. scVIDR predicts in vivo and in vitro dose-dependent gene expression across cell types in mouse liver, peripheral blood mononuclear cells, and cancer cell lines. We use regression to interpret the outputs of scVIDR. Additionally, we use scVIDR to order individual cells based on their sensitivities to a particular chemical by assigning a pseudo-dose value to each cell. Taken together, we show that scVIDR can effectively predict the dose and cell state dependent changes associated with chemical perturbations.