Benchmarking principal component analysis for large-scale single-cell RNA-sequencing.

Area of Science:

• Computational biology
• Genomics
• Bioinformatics

Background:

• Principal Component Analysis (PCA) is crucial for analyzing single-cell RNA-sequencing (scRNA-seq) data.
• Large-scale scRNA-seq datasets present computational challenges, including long processing times and high memory usage.

Purpose of the Study:

• To review and evaluate existing fast and memory-efficient PCA algorithms for scRNA-seq.
• To identify optimal PCA implementations for large-scale scRNA-seq data analysis.

Main Methods:

• Benchmarking of various PCA algorithms and implementations.
• Evaluation of computational performance (speed and memory efficiency) and accuracy.
• Focus on algorithms utilizing Krylov subspace and randomized singular value decomposition.

Main Results:

• Certain PCA algorithms, particularly those based on Krylov subspace and randomized SVD, demonstrate superior speed and memory efficiency.
• These selected algorithms maintain high accuracy when applied to large-scale scRNA-seq datasets.
• The study provides a comparative analysis of different PCA approaches.

Conclusions:

• Fast and memory-efficient PCA methods are available for large-scale scRNA-seq analysis.
• Krylov subspace and randomized SVD-based PCA algorithms are recommended for their performance.
• A guideline is developed to assist users in selecting the most appropriate PCA implementation based on their computational environment.