scSplit

scSplit demultiplexes pooled single-cell RNA-seq (scRNA-seq) data by inferring genetic differences directly from the sequencing data to assign cell barcodes to their original samples without requiring prior genotype information or external barcodes.

Key Features:

• Genotype-free demultiplexing: Identifies genetically distinct samples within pooled scRNA-seq data without requiring known sample genotypes or hashtag/barcode labels.
• Hidden state model: Uses a statistical hidden state model to infer genetic differences and separate mixed samples.
• Barcode assignment and cluster mapping: Assigns cell barcodes to clusters that map back to original samples to enable sample-specific analyses.
• Doublet annotation: Marks doublets using the notation DBL-<n>, where <n> indicates the cluster number.
• Concordance with existing methods: Reports high concordance with demuxlet results and with known truths in cell-hashing datasets.
• Empirical and simulated validation: Tested on real datasets of up to 8 mixed samples using the 10X Genomics pipeline and on simulated mixes of up to 32 samples.
• Output format: Produces CSV output detailing barcode-to-cluster (sample) assignments for downstream analysis.

Scientific Applications:

• Single-cell transcriptomics: Enables demultiplexing of pooled scRNA-seq experiments when sample genotypes or external identifiers are unavailable.
• Genetic diversity studies: Allows differentiation of closely related samples within mixed populations for genetic diversity analyses.
• Resource optimization: Facilitates pooling multiple samples per experiment to reduce sequencing cost and increase throughput.

Methodology:

Infers genetic differences from scRNA-seq data using a hidden state model, analyzes expression-derived variant signals to identify clusters corresponding to original samples, assigns barcodes to those clusters, and annotates doublets as DBL-<n>.