Imputing missing values in single-cell RNA-sequencing data: a statistical and machine learning-based approach

TL;DR

This paper introduces a new method called scDDI to better detect and fill in missing gene expression data in single-cell RNA sequencing.

Contribution

The novel scDDI method combines a Poisson–negative binomial mixture model with decision tree regression for improved dropout detection and imputation.

Findings

scDDI outperforms existing methods in dropout detection and imputation on both simulated and real datasets.

Improved imputation leads to better performance in downstream tasks like clustering and subpopulation identification.

Abstract

Single-cell RNA sequencing (scRNA-seq) offers a powerful tool to capture gene expression patterns within individual cells. However, due to the limited RNA content within cells, dropout events occur, resulting in a substantial number of zero counts in the single-cell expression matrix. To address this issue, we propose a novel method called single-cell dropout detection and imputation (scDDI). This method identifies dropout events using a Poisson–negative binomial mixture model and subsequently imputes the missing values using a decision tree regression model. We evaluate the performance of scDDI on both simulated and real scRNA-seq datasets, demonstrating its superiority over established single-cell imputation techniques. Notably, scDDI significantly improves dropout detection, leading to enhanced performance in various downstream analysis tasks like gene expression recovery, cell clustering, and cell subpopulation identification.

Figures