scReClassify

scReClassify refines post hoc cell type annotations in single-cell RNA sequencing (scRNA-seq) datasets using a semi-supervised adaSampling approach to identify and correct mislabeled cells.

Key Features:

• Semi-Supervised Learning Framework: Employs a semi-supervised approach that leverages both labeled and unlabeled scRNA-seq data to improve annotation accuracy.
• Dimension Reduction via PCA: Uses Principal Component Analysis (PCA) for dimension reduction to preserve key variance in complex scRNA-seq datasets prior to classification.
• Adaptive Sampling Algorithm (adaSampling): Implements the adaSampling algorithm to detect and reclassify mislabeled cells into their most probable cell types.
• Post hoc Reclassification: Performs post hoc refinement of initial cell type annotations to address mislabeling arising from incomplete biological knowledge or subjective inspection.
• Validation on Simulated and Experimental Data: Has been validated on both simulated and real-world experimental datasets across multiple tissues and biological systems.

Scientific Applications:

• Developmental Biology: Provides more accurate cell type classifications for studying cellular differentiation and developmental processes.
• Cell Reprogramming: Enhances precision in identifying reprogrammed cells for regenerative medicine and stem cell research.
• Cancer Research: Improves discrimination between cancerous and non-cancerous cell types to aid studies of tumor heterogeneity.

Methodology:

Applies Principal Component Analysis (PCA) for dimension reduction followed by a semi-supervised adaSampling algorithm that leverages labeled and unlabeled scRNA-seq data to identify and reclassify mislabeled cells; validation used simulated and real-world experimental datasets.