fastdemux: Robust SNP-based demultiplexing of single-cell population genomics data

Area of Science:

• Genomics
• Computational Biology
• Bioinformatics

Background:

• Sample multiplexing in single-cell genomics reduces costs and batch effects.
• Accurate and scalable computational demultiplexing is crucial for large-scale studies.
• Existing genotype-based methods like demuxlet can be computationally intensive.

Purpose of the Study:

• Introduce fastdemux, a novel computational framework for efficient genetic demultiplexing.
• Improve computational efficiency and scalability in demultiplexing pooled single-cell data.
• Maintain or enhance donor assignment accuracy.

Main Methods:

• Developed fastdemux based on a diagonal linear discriminant analysis (DLDA) model.
• Benchmarked fastdemux against demuxlet, vireo, and demuxalot using pooled single-cell RNA-seq data.
• Evaluated performance across varying sequencing depths and SNP filtering thresholds.

Main Results:

• Fastdemux demonstrated comparable or improved demultiplexing accuracy.
• Achieved orders of magnitude reduction in runtime and peak memory usage.
• DLDA framework successfully extended to doublet and multiplet detection.
• Effective performance observed with scATAC-seq data.

Conclusions:

• Fastdemux provides an efficient and scalable solution for genetic demultiplexing.
• Offers a significant computational advantage for large-scale single-cell genomics studies.
• Enables accurate donor assignment and multiplet detection in pooled datasets.