scASK

scASK implements an ensemble framework for classifying cell types from single-cell RNA-seq (scRNA-seq) data to handle high-dimensional and large-scale datasets for applications such as the Human Cell Atlas.

Key Features:

• Adaptive Slice KNNs framework: An ensemble architecture that integrates DAS, MCS, and EMS to improve classification performance on scRNA-seq data.
• DAS (Data Adaptive Slicing): Performs structural reduction by adaptively slicing the data to manage high dimensionality and improve computational efficiency.
• MCS (Meta Classifiers Selecting): Selects optimal classifiers from a pool of candidates to enhance classification performance.
• EMS (Ensemble Mode Switching): Dynamically switches between ensemble modes to achieve a bias–variance tradeoff and increase robustness.
• High performance on large-scale data: Designed to process large-scale scRNA-seq datasets and address poor performance and sensitivity to initial conditions seen in traditional methods.
• Versatility across data types: Applicable beyond scRNA-seq to supervised classification of gene expression and DNA methylation datasets.
• Empirical robustness and accuracy: Evaluated on thirteen real scRNA-seq datasets and reported to outperform five popular classification algorithms in accuracy and robustness.

Scientific Applications:

• Cell Type Identification: Precise identification and classification of cell types from complex scRNA-seq data, supporting efforts such as the Human Cell Atlas.
• Cancer Research: Classification of cancer subtypes using gene expression or DNA methylation profiles.
• General Supervised Classification Tasks: Use as a generic ensemble classifier for supervised learning across diverse biological and biomedical datasets.

Methodology:

Computational steps explicitly comprise Data Adaptive Slicing (DAS) for structural reduction, Meta Classifiers Selecting (MCS) for choosing classifiers, and Ensemble Mode Switching (EMS) for dynamic ensemble configuration within the Adaptive Slice KNNs ensemble framework.