CellOracle: Dissecting cell identity via network inference and in silico gene perturbation

Abstract

Here, we present CellOracle, a computational tool that integrates single-cell transcriptome and epigenome profiles to infer gene regulatory networks (GRNs), critical regulators of cell identity. Leveraging inferred GRNs, we simulate gene expression changes in response to transcription factor (TF) perturbation, enabling network configurations to be interrogated in silico, facilitating their interpretation. We validate the efficacy of CellOracle to recapitulate known regulatory changes across hematopoiesis, correctly predicting the outcomes of well-characterized TF perturbations. Integrating CellOracle analysis with lineage tracing of direct reprogramming reveals distinct network configurations underlying different reprogramming failure modes.Furthermore, analysis of GRN reconfiguration along successful reprogramming trajectories identifies new factors to enhance target cell yield, uncovering a role for the AP-1 subunit Fos, with the hippo signaling effector, Yap1. Together, these results demonstrate the efficacy of CellOracle to infer and interpret cell-type-specific GRN configurations, at high-resolution, promoting new mechanistic insights into the regulation and reprogramming of cell identity.

Learning interpretable cellular responses to complex perturbations in high-throughput screens

“…While powerful in interpreting interactions, mechanistic models usually require temporally-resolved data often unavailable in practice, and most do not scale to genome-wide measurements and challenges such as predicting high-dimensional scRNA-seq data. Linear models[12, 21] do not suffer from these scalability issues but have limited predictive power and are unable to capture non-linear cell-type specific responses. In contrast, deep learning (DL) models do not face these limitations.…”

scTenifoldKnk: an efficient virtual knockout tool for gene function predictions via single-cell gene regulatory network perturbation

“…Prediction of gene expression responses to perturbation using scRNA-seq data is an active research area. To the best of our knowledge, there are two software tools that have been developed for this purpose: scGen [3] and CellOracle [4]. scGen is a package implemented in Python, using TensorFlow variational autoencoders combined with vector arithmetic to predict gene expression changes in cells.…”

“…Also, essential genes cannot be knocked out. For that reason, predictive computational methods stand as a possible solution for such prohibitive experiments [2][3][4]. Taking advantage of the synchronized gene expression variability allowing the detection of co-expression between genes belonging to the same biological processes or being under the effect of the same transcription factors, it is now possible to construct personalized transcriptome-wide GRNs that are sample, tissue, and cell type-specific [5].…”

The Importance of Computational Modeling in Stem Cell Research

“…Indeed, scRNA-seq captures the gene expression of thousands of individual cells in one experiment, which provides a large number of independent measurements that allow the extraction of information about gene expression heterogeneity across cells and gene-gene coexpression in individual cells. Hence, scRNA-seq allows the inference of cell type-or cell subtype-specific GRNs [1][2][3][4][5][6][7][8][9][10][11], which constitutes an important step in the implementation of network-based methods for cellular conversion [12,13] (Table 1). In particular, using a machine learning approach, it has been possible to identify different GRN configurations corresponding to different states of pluripotency, including naive and primed states [4].…”