RAFSIL

RAFSIL learns cell–cell similarities from single-cell RNA sequencing (scRNA-seq) data to support accurate identification and characterization of cell types and subtypes.

Key Features:

• Random forest-based similarity learning: Uses random forest algorithms to learn cell–cell similarity measures from constructed features.
• scRNA-seq–tailored feature construction: Constructs features specifically optimized for scRNA-seq data to account for abundance and technical noise profiles.
• Two-step procedure: Implements a two-step workflow of feature construction followed by similarity learning.
• Technical variation detection: Identifies and highlights unwanted technical variation within scRNA-seq datasets.
• Integration for exploratory analyses: Provides similarity outputs suitable for downstream tasks such as dimension reduction, visualization, and clustering.
• Empirical performance: Demonstrates favorable performance compared to existing methods across diverse datasets.

Scientific Applications:

• Cell type identification: Facilitates identification and characterization of cell types and subtypes from scRNA-seq data.
• Dimension reduction and visualization: Supplies similarity measures usable for dimension reduction and visualization of cellular structure.
• Clustering and exploratory analysis: Supports clustering and other exploratory analyses of single-cell transcriptomic data.
• Data quality refinement: Aids detection of unwanted technical variation to improve data quality before downstream analyses.

Methodology:

RAFSIL performs a two-step computational procedure: it first constructs features optimized for scRNA-seq data and then applies similarity learning using random forest algorithms tailored to scRNA-seq abundance and technical noise.