scIntegral

scIntegral performs probabilistic cell-type identification and integration of multi-donor single-cell RNA sequencing (scRNA-seq) datasets to assign cell types while accounting for donor heterogeneity.

Key Features:

• Probabilistic Cell-Type Identification: Uses a semi-supervised approach that incorporates marker list information as prior knowledge to probabilistically assign cell types, reducing error rates by up to three-fold versus existing methods.
• Detection of Rare Cell Populations: Precisely identifies very rare cell populations at frequencies below 0.5%.
• Integration Across Multiple Donors: Integrates scRNA-seq data from multiple donors while accounting for strong heterogeneity and batch effects.
• Efficiency and Scalability: Scales to large cohorts, integrating datasets comprising ten thousand donors in approximately one hour while modeling heterogeneity with covariates.
• Quantitative Measurement of Integration Benefits: Quantifies integration benefits by comparing cell-type identification error rates against gold standard cell labels.

Scientific Applications:

• Cell-Type Identification: Accurate classification of cell types within complex scRNA-seq datasets to characterize cellular heterogeneity.
• Multi-Donor Data Integration: Integration of data from multiple donors to mitigate donor variability and batch effects in population-scale or multi-center studies.
• Rare Cell Population Analysis: Detection and analysis of very low-abundance cellular subtypes (<0.5%) for studies of novel cell types in disease or development.

Methodology:

Employs a semi-supervised learning framework that uses marker lists as priors for probabilistic cell-type assignment, models donor heterogeneity using covariates, and evaluates integration by comparing error rates to gold standard cell labels.