Denoising Single-Cell RNA-Seq Data with a Deep Learning-Embedded Statistical Framework

Abstract

Single-cell RNA sequencing (scRNA-seq) provides extensive opportunities to explore cellular heterogeneity, but is often limited by substantial technical noise and variability. To overcome these issues, we present ZILLNB, a novel computational framework that integrates zero-inflated negative binomial regression with deep generative modeling to simultaneously denoise and impute scRNA-seq data. By integrating a deep generative model, ZILLNB effectively captures latent group structures at both the cellular and gene levels, thereby enhancing the accuracy of recovered gene expression profiles while retaining cell-type-specific expression patterns. Comparative evaluations demonstrate that ZILLNB outperforms existing methods in accurately identifying differentially expressed genes and delineating biologically meaningful cell subpopulations. Moreover, ZILLNB exhibits versatile applicability, including effective correction of batch effects, making it a broadly useful tool for enhancing scRNA-seq data analyses.