K-Branches

K-Branches detects branching points in single-cell data by fitting local half-lines and selecting the number of branches to infer differentiation trajectories from single-cell RNA-Seq and single-cell qPCR datasets.

Key Features:

• Local half-line fitting: K-Branches locally fits half-lines to single-cell measurements to represent branches in cellular differentiation trajectories.
• K-Means-like clustering: It groups cells into branch-associated clusters using a clustering algorithm akin to K-Means.
• Model selection via modified GAP statistic: The method uses a modified GAP statistic to determine the optimal number of half-lines locally.
• Support for diverse single-cell data: The approach has been evaluated on single-cell RNA-Seq, single-cell qPCR, and artificial datasets including applications to myeloid progenitor differentiation, mouse blastocyst development, and human myeloid monocytic leukemia.

Scientific Applications:

• Hematopoiesis (myeloid progenitor differentiation): Detects branching events in single-cell RNA-Seq data from myeloid progenitor differentiation during hematopoiesis.
• Mouse blastocyst development (single-cell qPCR): Identifies branching structure in single-cell qPCR data from mouse blastocyst development.
• Human myeloid monocytic leukemia: Analyzes branching patterns in single-cell data from human myeloid monocytic leukemia.
• Method benchmarking with artificial data: Evaluates method behavior and performance on artificial datasets.

Methodology:

Inputs are single-cell RNA-Seq or single-cell qPCR data; the method fits local half-lines to identify candidate branches, groups cells using a K-Means–like clustering algorithm, and selects the number of half-lines via a modified GAP statistic.