scHPL

scHPL implements hierarchical progressive learning to integrate multiple single-cell datasets annotated at varying resolutions into a hierarchical classifier that identifies known and unknown cell populations.

Key Features:

• Hierarchical Classification Tree Construction: Automatically constructs a hierarchical classification tree by discovering relationships between cell populations across multiple datasets.
• Support for Multiple Datasets: Learns from several single-cell datasets simultaneously and handles differing annotation resolutions.
• Continuous/Progressive Learning: Updates the hierarchical classifier as new data are incorporated to refine the learned structure.
• Detection of Unknown Populations: Applies a linear Support Vector Machine (SVM) or a one-class SVM at each node to classify cells and detect previously unannotated populations.
• Preservation of Annotations: Maintains original annotations during retraining to ensure consistency across updates.

Scientific Applications:

• Cell type identification: Identifying and classifying cell types within complex single-cell biological samples.
• Cellular hierarchy and lineage analysis: Resolving hierarchical relationships and lineage structure across conditions or developmental stages.
• Cross-dataset integration: Integrating diverse single-cell datasets to produce a unified view of cellular diversity.
• Validation and benchmarking: Demonstrating method performance on simulated and real datasets for evaluating hierarchical classification fidelity.

Methodology:

scHPL constructs hierarchical classification trees from multiple datasets annotated at varying resolutions, applies linear SVMs or one-class SVMs at each node, and uses a progressive learning procedure that updates the tree while preserving original annotations.