NeuCA

NeuCA classifies single-cell RNA sequencing (scRNA-seq) data using a neural-network framework that incorporates hierarchical cell-type structure to improve cell-type annotation accuracy.

Key Features:

• Neural-network based classification: Uses a neural-network model to assign cell-type labels to scRNA-seq profiles.
• Hierarchical cell-type modeling: Incorporates the hierarchical structure of cell types into the classification framework.
• Dynamic classification strategy: Adjusts classification decisions based on cell-type correlations to improve discrimination of similar types.
• Supervised label assignment: Leverages existing scRNA-seq reference databases for supervised annotation.
• Mitigates "unassigned" labels: Reduces the tendency to label low-confidence cells as "unassigned" compared with some existing methods.
• Performance on correlated cell types: Specifically improves annotation accuracy in datasets containing closely correlated or highly similar cell types.
• Validation across datasets: Demonstrated and validated on eight real scRNA-seq datasets.
• Robust intra-/inter-study and cross-condition annotation: Shows stable intra- and inter-study transfer and cross-condition annotation capability.

Scientific Applications:

• Cell-type annotation in heterogeneous tissues: Assigns cell-type labels in heterogeneous single-cell transcriptomic datasets.
• Discrimination of closely related cell types: Improves identification of highly similar or correlated cell populations.
• Cross-study annotation transfer: Enables annotation transfer across studies (intra- and inter-study).
• Cross-condition annotation: Facilitates annotation across different experimental conditions.
• Benchmarking and comparative evaluation: Serves in comparative analyses to evaluate annotation accuracy relative to existing methods.

Methodology:

Implements a supervised neural-network classifier that incorporates hierarchical cell-type structure and leverages existing scRNA-seq reference databases for label assignment.