CITE-sort

CITE-sort performs artificial-cell-type-aware clustering of antibody-derived tag (ADT) data from Cellular Indexing of Transcriptomes and Epitopes by sequencing (CITE-seq) to improve surface marker clustering and separate biological cell types (BCTs) from multiplet-induced artificial cell types (ACTs).

Key Features:

• Artificial-cell-type-aware clustering: Incorporates awareness of multiplets and artificial cell types (ACTs) when clustering CITE-seq ADT data.
• Recursive Gaussian Mixture Model: Uses a recursive Gaussian Mixture Model to partition ADT signal distributions.
• Auto-gating on ADT data: Performs automated gating of antibody-derived tag signals to define clusters.
• Distinguishes BCTs and ACTs: Separates real biological cell types (BCTs) from artificial clusters generated by multiplets.
• Binary tree clustering structure: Organizes clustering results into a binary tree to represent recursive partitioning.
• Supports cell-type annotation: Facilitates annotation by structuring clusters in a way that incorporates CITE-seq domain knowledge.
• Robustness to multiplets: Maintains clustering accuracy in the presence of multiplet-induced artifacts.
• Validated benchmarking: Demonstrated superior clustering performance on both real and simulated CITE-seq datasets compared to canonical methods.

Scientific Applications:

• Surface marker clustering: Improves accuracy of surface marker-based cell type identification in CITE-seq experiments.
• Multiplet separation: Identifies and separates multiplet-induced artificial cell-type clusters from true single-cell profiles.
• Cell-type annotation: Aids manual or automated cell-type annotation by providing an interpretable cluster hierarchy.
• Method benchmarking: Serves as a benchmark method for evaluating clustering performance on real and simulated CITE-seq datasets.

Methodology:

Applies recursive Gaussian Mixture Model fitting as an auto-gating procedure on ADT data to produce a binary tree of clusters that separate biological cell types (BCTs) from artificial cell types (ACTs), and evaluates performance on real and simulated CITE-seq datasets.