Accelerated dimensionality reduction of single-cell RNA sequencing data with fastglmpca.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Principal Components Analysis (PCA) is widely used for dimensionality reduction.
• Applying standard PCA to count data, such as single-cell RNA sequencing (scRNA-seq) data, presents theoretical and practical challenges.
• Existing methods like Poisson GLM-PCA are computationally intensive.

Purpose of the Study:

• To address the computational challenges of fitting Poisson Generalized Linear Model PCA (GLM-PCA).
• To introduce a more efficient and effective algorithm for dimensionality reduction of scRNA-seq count data.
• To provide an R package, fastglmpca, for implementing the new algorithm.

Main Methods:

• Developed and implemented a novel algorithm named Alternating Poisson Regression (APR).
• APR is designed to be memory-efficient and amenable to parallel processing on multi-core systems.
• Evaluated APR's performance against existing algorithms using two published scRNA-seq datasets.

Main Results:

• APR achieves better quality fits than existing GLM-PCA algorithms.
• APR demonstrates significantly reduced computation time compared to current methods.
• The algorithm is memory-efficient, facilitating analysis of large scRNA-seq datasets.

Conclusions:

• Alternating Poisson Regression (APR) provides a computationally efficient and accurate alternative for dimensionality reduction in scRNA-seq data.
• The fastglmpca R package offers a practical implementation of APR, aiding researchers in analyzing large-scale count data.
• This advancement can improve the scalability and accessibility of PCA-based methods for scRNA-seq data analysis.