scAnnotate

scAnnotate annotates cell types in single-cell RNA-sequencing (scRNA-seq) data by modeling dropout events and non-dropout expression distributions to improve cell-type classification.

Key Features:

• Dropout Information Utilization: Explicitly models dropout events inherent to scRNA-seq data to account for genes that are undetected in individual cells.
• Mixture Model Approach: Uses a mixture model for the marginal distribution of each gene to capture both the dropout proportion and the distribution of non-dropout expression levels.
• Ensemble Machine Learning Strategy: Combines gene-wise mixture models via an ensemble machine learning approach to produce a cohesive cell-type annotation model and reduce parameter estimation complexity in the high-dimensional joint distribution.
• Empirical Performance Evaluation: Assessed on 14 real scRNA-seq datasets and compared against nine existing annotation methods, yielding a distinct pattern of misclassified cells relative to other tools.

Scientific Applications:

• Cellular heterogeneity characterization: Provides cell-type labels to dissect cellular heterogeneity within biological samples at single-cell resolution.
• Downstream single-cell analyses: Supports downstream analyses such as differential expression analysis and functional studies at the single-cell level.

Methodology:

Fits gene-wise mixture models that model dropout proportion and non-dropout expression distributions, combines these models across genes via an ensemble machine learning strategy to assign cell types, and evaluates performance on 14 real scRNA-seq datasets against nine existing annotation methods.