scHiCStackL

scHiCStackL employs a two-layer stacking ensemble to classify cell types from single-cell Hi-C data, improving prediction accuracy through refined preprocessing and representative cell embeddings.

Key Features:

• Refined data preprocessing: Implements an improved preprocessing method for single-cell Hi-C data that yields more representative cell embeddings.
• Cell embeddings: Produces refined cell embeddings that serve as the foundation for classification.
• Two-layer stacking ensemble model: Uses a two-layer stacking ensemble specifically constructed for cell-type classification from single-cell Hi-C profiles.
• Benchmarking against scHiCluster: Demonstrates performance improvements over scHiCluster with increases of 13.33%, 19%, 19.27%, and 14.5% in Accuracy (Acc), Adjusted Rand Index (ARI), Normalized Mutual Information (NMI), and F1 score confidence intervals, respectively, on human datasets ML1 and ML3.
• Preprocessing quantitative gains: Reports preprocessing-related improvements of 0.23%, 1.22%, 1.46%, and 1.61% in Acc, Matthews Correlation Coefficient (MCC), F1 score, and Precision confidence intervals compared to scHiCluster.

Scientific Applications:

• Cell-type classification from single-cell Hi-C: Enables distinguishing cell types based on single-cell Hi-C three-dimensional chromosomal interaction profiles.
• Benchmarking on human datasets: Applied to human cell datasets ML1 and ML3 for comparative performance evaluation against existing methods such as scHiCluster.

Methodology:

Refined single-cell Hi-C data preprocessing; generation of representative cell embeddings; a two-layer stacking ensemble model for classification; evaluation on human datasets ML1 and ML3 using Accuracy (Acc), Adjusted Rand Index (ARI), Normalized Mutual Information (NMI), F1 score, Matthews Correlation Coefficient (MCC), and Precision.