Dimensionality reduction for single cell RNA sequencing data using constrained robust non-negative matrix factorization | JoVE Visualize

Area of Science:

Genomics
Bioinformatics
Computational Biology

Background:

Single-cell RNA sequencing (scRNA-seq) is a powerful technology for transcriptome analysis at the individual cell level.
scRNA-seq data is characterized by 'dropouts' (missing gene expression values), which can negatively impact downstream analyses like clustering and dimensionality reduction.
Dropout rates are inversely correlated with sequencing depth, with lower depths leading to more dropouts.

Purpose of the Study:

To develop a novel dimensionality reduction method specifically designed for scRNA-seq data.
To address the challenges posed by data dropouts and non-negativity constraints inherent in scRNA-seq datasets.
To improve the accuracy and robustness of downstream analyses by integrating dropout imputation with dimensionality reduction.

Main Methods:

A new dimensionality reduction technique was developed within the non-negative matrix factorization (NMF) framework.
The method simultaneously performs dimensionality reduction and dropout imputation.
Dropouts were modeled as a non-negative sparse matrix, and NMF was used to approximate the sum of observed and dropout matrices.
A weighted ℓ1 penalty was incorporated to maintain sparsity, accounting for the relationship between dropout rates and sequencing depth per cell.
An efficient algorithm was devised to solve the optimization problem.

Main Results:

The proposed method demonstrated more robust clustering results on both synthetic and real scRNA-seq data compared to existing methods.
Dimensionality reduction using the developed technique improved the performance of clustering algorithms.
Dropout imputation integrated into the method enhanced the accuracy of differential gene expression analysis.

Conclusions:

The developed NMF-based dimensionality reduction method effectively handles dropouts in scRNA-seq data.
The approach leads to improved clustering and differential expression analysis, offering a more reliable tool for scRNA-seq data interpretation.
This method provides a valuable advancement for researchers utilizing scRNA-seq technology.