FSCAM

FSCAM performs feature selection using convex analysis of mixtures to improve clustering and cell-type identification from single-cell RNA sequencing (scRNA-seq) data.

Key Features:

• Advanced Feature Selection: Uses convex analysis of mixtures to select features by evaluating gene relevancy, redundancy, and completeness while modeling many-to-many gene interactions.
• Improved Clustering Accuracy: Integrates selected features with Partition Around Medoids (PAM) to implement SCC_FSCAM, enhancing accuracy of cell type determination from scRNA-seq data.
• Benchmarking and Validation: Benchmarking on real datasets evaluated internal criteria (optimal clustering number) and external criteria (adjusted Rand index) and demonstrated improved stability and performance versus traditional methods.

Scientific Applications:

• Cell type determination: Improves identification of cell types from transcriptome profiles generated by scRNA-seq.
• Analysis of cellular heterogeneity: Facilitates characterization of cellular heterogeneity across biological contexts through more discriminative features and stable clustering.

Methodology:

Applies convex analysis of mixtures for feature selection by assessing gene relevancy, redundancy, and completeness and modeling many-to-many gene interactions, integrates selected features with Partition Around Medoids (PAM) to form SCC_FSCAM, and evaluates performance using internal criteria (optimal clustering number) and external criteria (adjusted Rand index).