Sparsity-Penalized Stacked Denoising Autoencoders for Imputing Single-Cell RNA-seq Data

Abstract

Single-cell RNA-seq (scRNA-seq) is quite prevalent in studying transcriptomes, but it suffers from excessive zeros, some of which are true, but others are false. False zeros, which can be seen as missing data, obstruct the downstream analysis of single-cell RNA-seq data. How to distinguish true zeros from false ones is the key point of this problem. Here, we propose sparsity-penalized stacked denoising autoencoders (scSDAEs) to impute scRNA-seq data. scSDAEs adopt stacked denoising autoencoders with a sparsity penalty, as well as a layer-wise pretraining procedure to improve model fitting. scSDAEs can capture nonlinear relationships among the data and incorporate information about the observed zeros. We tested the imputation efficiency of scSDAEs on recovering the true values of gene expression and helping downstream analysis. First, we show that scSDAE can recover the true values and the sample–sample correlations of bulk sequencing data with simulated noise. Next, we demonstrate that scSDAEs accurately impute RNA mixture dataset with different dilutions, spike-in RNA concentrations affected by technical zeros, and improves the consistency of RNA and protein levels in CITE-seq data. Finally, we show that scSDAEs can help downstream clustering analysis. In this study, we develop a deep learning-based method, scSDAE, to impute single-cell RNA-seq affected by technical zeros. Furthermore, we show that scSDAEs can recover the true values, to some extent, and help downstream analysis.

Guidelines for bioinformatics of single-cell sequencing data analysis in Alzheimer’s disease: review, recommendation, implementation and application

“…Deep neural networks consist of several layers of encoders mapping the input data into a low-dimensional manifold, from which the following decoder layers can reconstruct denoised, full-rank data. The applications in scRNA-seq include denoising single-cell transcriptome [ 95 , 122 ], batch effect removal [ 118 ], probabilistic modeling of gene expressions or cell types [ 95 , 96 , 123 ] or dimension reduction [ 96 , 117 , 118 , 124 ]. In cell clustering, these versatile functions of deep neural networks have become an attractive avenue to unveil complex cell architectures in scRNA-seq.…”

“…Classically, compared to the original input, these deep neural network models are trained by minimizing the reconstructed data loss. However, naïve model learning in this way could lead to over-fitting where non-biological sources of errors (e.g., drop-out reads, low coverage) in scRNA-seq contribute differently to the data noises [ 122 ]. Deep count autoencoder (DCA) and single-cell variational inference (scVI) define the reconstruction error as the log-likelihood of the noise model such as ZINB to denoise and impute the drop-out reads.…”

Application of Deep Learning on Single-Cell RNA Sequencing Data Analysis: A Review

“…However, they are usually computationally intensive and limited in capturing non-linearity in scRNA-seq data. To better address this issue, deep learning approaches have been developed for scRNA-seq data imputation and denoising [37] , [38] , [39] , [40] , [41] , [42] , [43] , [44] , [45] , [46] , [47] , [48] , [49] . Based on an idea similar to regression imputation [50] , i.e.…”

“…To overcome this, ZINB model-based autoencoder (ZINBAE) [42] developed a ZINB autoencoder by introducing a differentiable function [51] to approximate the categorical data and a regularization term to control the ZINB. Sparsity-penalized stacked denoising autoencoder (scSDAE) [43] leveraged a stacked DAE for scRNA-seq imputation with L1 loss to prevent overfitting. GraphSCI [44] combined the graph convolutional network (GCN), a type of GNN, with the standard autoencoder to model gene–gene co-expression relations and single-cell gene expression matrix, respectively.…”

Understanding the pathogenesis of infectious diseases by single-cell RNA sequencing

“…This limit may result in artificial zeros either systematically or accidently, thus hindering complete downstream analysis [ 81 ]. Using computational statistical models, it may be possible to circumvent this limitation by modulation of sample variation and noise by use of VIPER [ 82 ] and scSDAEs [ 83 ]. With regard to infection, most studies have focused on either the host or the pathogen, whereas dual RNA-seq allows analysis of the transcripts of both host and bacteria simultaneously [ 84 ], but not at the single-cell bacterial level.…”