A scalable approach to topic modelling in single-cell data by approximate pseudobulk projection.

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Probabilistic topic modeling is crucial for single-cell data analysis, enabling the identification of latent cellular states.
• Current topic modeling methods demand substantial computational resources (time, memory, specialized hardware) for large datasets.
• Interpretable bases for comparison with known markers and pathways are derived from topic-specific gene frequency vectors.

Purpose of the Study:

• To introduce a scalable approximation method for single-cell RNA-seq data analysis.
• To address the computational limitations of existing topic modeling techniques for large-scale single-cell datasets.
• To provide an accurate and efficient alternative for identifying cellular states and integrating diverse data types.

Main Methods:

• Developed ASAP (Annotating a Single-cell data matrix by Approximate Pseudobulk estimation), a novel scalable approximation method.
• Customized the method specifically for single-cell RNA-seq data analysis.
• Demonstrated seamless integration of single-cell and bulk data without additional preprocessing or feature selection.

Main Results:

• ASAP achieves higher accuracy compared to existing methods.
• The method requires significantly less computing time (orders of magnitude) and lower memory consumption.
• ASAP is effective for atlas-scale data analysis and joint analysis of single-cell and bulk data.

Conclusions:

• ASAP offers a computationally efficient and accurate solution for probabilistic topic modeling in single-cell genomics.
• The method overcomes the resource-intensive nature of traditional approaches, enabling large-scale analyses.
• ASAP facilitates broader applicability in single-cell data analysis, including the integration of heterogeneous data sources.