scSimClassify

scSimClassify performs reference-free, alignment-free multi-class cell type classification from raw scRNA-seq reads by extracting k-mer features and grouping them with simhash into Compressed K-mer Groups (CKGs) to improve utilization of unmapped reads and classification accuracy.

Key Features:

• Reference-free and alignment-free: Operates directly on raw scRNA-seq reads without mapping to reference genomes or performing gene quantification.
• K-mer level feature extraction: Extracts k-mers from raw sequencing reads as fundamental features to capture sequence-level signals that may be lost in gene expression quantification.
• Simhash compression into Compressed K-mer Groups (CKGs): Uses the simhash method to group k-mers with similar abundance profiles into CKGs, reducing dimensionality and grouping correlated signals.
• Improved classification accuracy: CKG features outperform traditional gene expression features for multi-class cell type classification across four scRNA-seq datasets.
• Retention of unmapped reads and reduced alignment overhead: By bypassing alignment, retains information from unmapped reads and reduces reliance on resource-intensive mapping steps.

Scientific Applications:

• Multi-class cell type classification: Identification of cell types in scRNA-seq datasets using k-mer and CKG features.
• Analysis of sparse or low-depth scRNA-seq data: Application to datasets with limited sequencing depth per cell or inherent data sparsity by leveraging sequence-level signals.
• Inclusion of unmapped reads in analyses: Enables use of unmapped raw reads for classification and downstream biological interpretation.
• Genomics and translational studies: Supports genomics and personalized medicine research that requires robust cell population delineation.
• Feature comparison and method benchmarking: Facilitates comparison of k-mer/CKG features with gene expression features and other state-of-the-art approaches.

Methodology:

Extract k-mers from raw scRNA-seq reads; apply the simhash method to group similar k-mers into Compressed K-mer Groups (CKGs); use CKG features for multi-class cell type classification and comparison with state-of-the-art methods and scRNA-seq specific algorithms.