An Efficient and Reliable scRNA-seq Data Imputation Method Using Variational Autoencoders

Single-cell RNA sequencing (scRNA-seq) provides the expression profiles of individual cells to study cell-to-cell variation within a cell population and analyses single-cell RNA-seq data to discover population heterogeneity. However, the original scRNA-seq data often contains many zeros due to dropout events, which can cause significant deviation during analysis. Although deep learning-based methods have been developed for imputing missing data in scRNA-seq, they are not explicitly designed to model the data distribution, hindering the generation of multiple plausible imputations. In this study, we propose a novel method to restore incomplete scRNA-seq data by handling dropouts and retaining true zeros with potential applications in Cancer research projects. Our method employs a variational autoencoder (VAE) as a deep generative model to learn the data distribution and reconstruct the imputed cell-gene matrix. VAE can learn a low-dimensional representation of the data that captures the most important features to generate plausible imputations. Several experiments have been conducted to evaluate the accuracy and efficiency of our method compared to state-of-the-art methods using datasets of various sizes. Evaluation metrics such as mean squared error (MSE) and clustering tests were employed, and the results demonstrated that our method outperformed other approaches. Interestingly, our proposed model exhibit strong stability when dealing with different data magnitudes.