scPred

scPred classifies cells from single-cell RNA sequencing (scRNA-seq) data by extracting informative principal components via singular value decomposition and applying machine learning to identify cell-type transcriptional signatures.

Key Features:

• Variance Structure Decomposition: Uses singular value decomposition and principal component analysis to identify principal components that capture variance informative of cell type and reduce dimensionality.
• Unbiased Feature Selection: Selects features from the reduced-dimension space to retain only the most relevant components for classification.
• Machine Learning Classification: Trains probability-based machine learning models on selected principal components to predict cell types.
• Generalizability Across Datasets: Applies the same variance-driven dimensionality reduction and classifier framework across diverse scRNA-seq datasets and biological contexts.

Scientific Applications:

• Cell Type Identification: Classifies individual cells by transcriptional profile to characterize specific cell types within complex tissues.
• Transcriptional Signature Analysis: Identifies defining transcriptional signatures for studies of cellular differentiation and state transitions.
• Cross-tissue and Condition Studies: Applied to diverse datasets including pancreatic tissue, mononuclear cells, colorectal tumor biopsies, and circulating dendritic cells.

Methodology:

Data preprocessing; variance-structure decomposition via singular value decomposition/principal component analysis; unbiased feature selection from the reduced-dimension space; machine learning model training on selected PCs; probability-based prediction and classification of cells in new datasets.