A fast and efficient count-based matrix factorization method for detecting cell types from single-cell RNAseq data

Area of Science:

• Computational biology
• Genomics
• Bioinformatics

Background:

• Single-cell RNA sequencing (scRNAseq) data analysis faces challenges with unwanted variables masking true cell-type signals.
• Existing matrix factorization tools for scRNAseq data often fail to model raw counts or are too slow for large datasets (n>500).

Purpose of the Study:

• To develop a fast, efficient, and count-based matrix factorization method for inferring cell-type structure from scRNAseq data.
• To address the limitations of existing methods in handling large-scale scRNAseq datasets.

Main Methods:

• Developed single-cell negative binomial matrix factorization (scNBMF), a count-based matrix factorization method.
• Utilized the TensorFlow framework for efficient computation.
• Validated the method on three public scRNAseq datasets: brain, embryonic stem, and pancreatic islet.

Main Results:

• scNBMF effectively infers low-dimensional structure for cell-type identification.
• Experimental results demonstrate scNBMF's superior power in detecting cell types compared to existing bespoke tools.
• scNBMF achieves 10-100 fold speed improvement on large scRNAseq datasets.

Conclusions:

• scNBMF is a powerful and efficient tool for large-scale scRNAseq data analysis, particularly for cell-type detection.
• The method is implemented in R and Python, with source code publicly available.
• scNBMF offers a scalable solution for analyzing complex single-cell gene expression data.