Matrix prior for data transfer between single cell data types in latent Dirichlet allocation.

Area of Science:

• Genomics
• Computational Biology
• Molecular Biology

Background:

• Single cell ATAC-seq (scATAC-seq) is a powerful technique for mapping regulatory elements in individual cells.
• Analyzing scATAC-seq data is complex, and generating large datasets is costly and challenging.
• Leveraging existing scATAC-seq or scRNA-seq data can aid the analysis of new datasets.

Purpose of the Study:

• To investigate whether nonuniform matrix priors, derived from existing datasets, can improve cell type detection in new scATAC-seq data.
• To assess the efficacy of this approach, particularly for smaller scATAC-seq datasets.

Main Methods:

• Applied latent Dirichlet allocation (LDA), a Bayesian algorithm, to analyze scATAC-seq data, treating cells as documents and accessible sites as words.
• Compared standard uniform symmetric priors with nonuniform matrix priors generated from pre-trained LDA models.
• Tested the approach using scATAC-seq data from C. elegans nematodes and SHARE-seq data from mouse skin cells.

Main Results:

• Nonsymmetric matrix priors significantly improved the ability to capture cell type information.
• The enhancement was particularly notable in smaller scATAC-seq datasets.
• This demonstrates the utility of leveraging prior data through informed priors in LDA for scATAC-seq analysis.

Conclusions:

• Nonuniform matrix priors offer a significant advantage for cell type identification in scATAC-seq data analysis.
• This method enhances the interpretability of scATAC-seq data, especially when dealing with limited sample sizes.
• The findings provide a more robust framework for analyzing single-cell epigenomic data.