BiGCN: Leveraging Cell and Gene Similarities for Single-cell Transcriptome Imputation with Bi-Graph Convolutional Networks

Abstract

MotivationRNA profiling at the single-cell level is essential for characterizing the molecular activities and functions of individual cells. The current technical limitations of single-cell RNA sequencing (scRNA-seq) technologies can lead to a phenomenon known as "dropout", where a significant portion of gene expression is not captured. Dropout is particularly prominent in genes with low or sparse expression, greatly impacting the reliability and interpretability of scRNA-seq data. Consequently, various techniques have been developed to estimate missing gene expression using imputation, often by either modeling similarities in gene expression among cells or using gene co-expression, but rarely both. ResultsIn this study, we introduce a Bi-Graph Convolutional Network (BiGCN), a deep learning method that leverages both cell similarities and gene co-expression to capture cell-type-specific gene co-expression patterns for imputing scRNA-seq data. BiGCN constructs both a cell similarity graph and a gene co-expression graph, and employs them for convolutional smoothing in a dual two-layer Graph Convolutional Networks (GCNs). The embeddings from the two GCNs can subsequently be combined to facilitate the final imputation. BiGCN demonstrates superior performance compared to state-of-the-art imputation methods on both real and simulated scRNA-seq data. Additionally, BiGCN outperforms existing methods when tasked with clustering cells into cell types. We also perform a novel validation using a PBMC scRNA-seq dataset, and this experiment supports that BiGCNs imputations are more realistic than competing imputation methods. In both the imputation and the cluster tasks, BiGCN consistently outperformed two variants of BiGCN that solely relied on either the gene co-expression graph or cell similarity graph. This indicates that the two graphs offer complimentary information for imputation and cell clustering, underscoring the importance of incorporating both types of information. Code Availabilityhttps://github.com/inoue0426/scBiGCN. Contactkuang@umn.edu