scRFE

scRFE identifies minimal gene sets that define cellular identities from single-cell RNA sequencing (scRNA-seq) data using a one-versus-all random forest classifier combined with recursive feature elimination.

Key Features:

• Random Forest Classifier: Uses a random forest algorithm to classify cells based on gene expression profiles.
• One-vs-All Classification: Performs one-versus-all classification to distinguish each cell type or state from all others.
• Recursive Feature Elimination (RFE): Applies recursive feature elimination to iteratively remove less important genes and retain a minimal feature set.
• Cross-Validation: Incorporates cross-validation during feature selection to assess robustness and reduce overfitting.
• Feature Importance Ranking: Ranks features by importance to identify critical genes and minimal transcriptional programs.
• Scanpy Integration: Compatible with Scanpy for integration into Scanpy-based single-cell analysis workflows.
• High-Dimensional Data Handling: Handles the high-dimensional nature of scRNA-seq gene expression data.

Scientific Applications:

• Cell classification by metadata: Generates interpretable gene lists that classify cells according to metadata categories within scRNA-seq datasets.
• Minimal transcriptional programs: Identifies minimal and ranked gene sets that describe biological features of a dataset.
• Regulatory network inference support: Highlights key genes that facilitate identification of regulatory networks and molecular determinants of cellular identity.
• Aging analysis (Tabula Muris Senis): Has been applied to the Tabula Muris Senis dataset to reproduce established aging patterns.
• Transcription factor reprogramming: Has been used to elucidate transcription factor reprogramming protocols.

Methodology:

Implements a one-versus-all random forest classifier combined with recursive feature elimination and cross-validation to rank feature importance and select minimal gene sets, and is compatible with Scanpy-based workflows.