CTISL

CTISL applies a two-layer stacking ensemble learning model to identify cell types from single-cell RNA sequencing (scRNA-seq) data, improving classification accuracy for studies of cellular heterogeneity and function.

Key Features:

• Ensemble learning strategy: Implements a two-layer stacking model that integrates multiple classifiers to enhance prediction performance beyond single classifiers.
• Dynamic integration of classifiers: Uses a reference scRNA-seq dataset with known cell types in the first layer to train and dynamically combine cell-type-specific base learners, including support-vector machines and logistic regression.
• Meta-classification: Feeds outcomes from base learners into a meta-classifier in the second layer to optimize overall cell type assignment.
• Benchmarking and validation: Evaluated across 24 benchmarking experiments using 17 diverse human and mouse scRNA-seq datasets to compare performance against existing predictors.

Scientific Applications:

• Developmental biology: Enables identification of cell types and states across developmental time courses using scRNA-seq.
• Disease pathology: Supports mapping of altered cell-type compositions and states in disease contexts.
• Immunology: Facilitates classification of immune cell populations and their heterogeneity in single-cell datasets.
• Regenerative medicine: Assists in characterizing cell types relevant to tissue repair and stem cell therapies.
• Cellular diversity and function analysis: Improves resolution of cellular heterogeneity to inform studies of function and lineage relationships.

Methodology:

Constructs a two-layer stacking model that trains cell-type-specific base learners (e.g., support-vector machines and logistic regression) on a reference scRNA-seq dataset with known cell types in the first layer, uses base-learner outputs as inputs to a meta-classifier in the second layer, and assesses performance via 24 benchmarking experiments across 17 human and mouse scRNA-seq datasets.