Deep learning for inferring gene relationships from single-cell expression data

Abstract

Several methods were developed to mine gene–gene relationships from expression data. Examples include correlation and mutual information methods for coexpression analysis, clustering and undirected graphical models for functional assignments, and directed graphical models for pathway reconstruction. Using an encoding for gene expression data, followed by deep neural networks analysis, we present a framework that can successfully address all of these diverse tasks. We show that our method, convolutional neural network for coexpression (CNNC), improves upon prior methods in tasks ranging from predicting transcription factor targets to identifying disease-related genes to causality inference. CNNC’s encoding provides insights about some of the decisions it makes and their biological basis. CNNC is flexible and can easily be extended to integrate additional types of genomics data, leading to further improvements in its performance.

Noise regularization removes correlation artifacts in single-cell RNA-seq data preprocessing

“…Pioneering work by Iacono et al used single-cell data-derived correlation metrics to generate gene regulatory networks and found that the networks could detect latent regulatory changes 30 . A deep learning approach has also been developed to predict transcription factor targets from single-cell expression data 31 . In this study, we used single cell gene-gene correlations derived after noise regularization to reconstruct a gene network that produced clear immune cell type related modules.…”

Deep learning of gene relationships from single cell time-course expression data

“…We relied on a strict peak p-value cutoff to identify binding sites (p value < 10 -400 based on MACS2 (Zhang, et al, 2008)). We next defined a 'promoter region' as 10Kb upstream and 1Kb downstream from the TSS of each gene and assigned TFs to regulate genes if a peak for that TF was identified in the promotor region for the gene as has been previously done (Schulz, et al, 2013;Yuan and Bar-Joseph, 2019).…”

“…The output of all models is a N-dimension (ND) vector, where N depends on specific task on which the TDL is trained (for example, N=1 for interaction predictions and 3 for interaction and causality predictions). The first TDL model we consider is termed 3D CNN and is a direct extension of the 2D CNNC method (Yuan and Bar-Joseph, 2019). In general, 3D CNN consists of one T×8×8 tensor input layer, several intermediate 3D convolutional…”

“…We compared the TDL approaches (3D CNN and Conv-LSTM) with a 2D CNNC (Yuan and Bar-Joseph, 2019) (which is also supervised), and unsupervised methods that utilize granger causality and lagged regression analysis to infer the interaction direction (Kim, et al, 2011). The granger causality inference method was applied to 3 of the 4 datasets since the 2 nd (mESC2) only had 4 time points which turned out to be less than the minimal number of time points required by the method.…”

“…We have recently developed the convolutional neural network for coexpression (CNNC) analysis method which is focused on inferring gene-gene relationships from static scRNA-Seq data (Yuan and Bar-Joseph, 2019). Unlike prior methods for inferring interactions, CNNC does not make any assumptions about the specific attributes or correlations that can be observed for interacting genes.…”