scIRT: Imputation and Dimensionality Reduction for Single-Cell RNA-Seq Data by Combining NMF with SMOTE

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Single-cell RNA sequencing (scRNA-seq) enables high-resolution analysis of cellular gene expression.
• Experimental scRNA-seq data suffers from high dropout rates, resulting in incomplete gene expression matrices.
• Existing imputation methods often focus on high-dimensional matrices and do not provide effective low-dimensional representations for clustering.

Purpose of the Study:

• To develop a novel computational tool for imputing missing data and performing dimensionality reduction on scRNA-seq data simultaneously.
• To address the limitations of current imputation techniques by integrating data imputation with dimensionality reduction.
• To enhance the accuracy and robustness of downstream analyses, such as cell type clustering and visualization.

Main Methods:

• Designed scIRT, an iterative imputation pipeline combining Synthetic Minority Over-sampling Technique (SMOTE) and Non-negative Matrix Factorization (NMF).
• SMOTE was adapted to effectively impute missing gene expression values (dropout events).
• NMF was employed for dimensionality reduction and feature extraction from the high-dimensional scRNA-seq data.

Main Results:

• The scIRT pipeline demonstrated superior and more robust performance compared to existing methods across multiple scRNA-seq datasets.
• Evaluated the impact of both the imputed matrix and the low-dimensional representation matrix on clustering accuracy.
• Showcased scIRT's capability to effectively recover missing data and facilitate downstream analyses like cell type clustering and visualization.

Conclusions:

• scIRT is an effective tool for preprocessing scRNA-seq data, addressing critical challenges of missing data and dimensionality.
• The integrated approach of imputation and dimensionality reduction provides a more comprehensive representation for biological interpretation.
• scIRT facilitates improved cell type identification and visualization, advancing the utility of scRNA-seq data in biological research.