Model-based branching point detection in single-cell data by K-branches clustering.

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Single-cell technologies like single-cell RNA-Seq reveal cell population heterogeneities, enabling inference of cellular development and lineage trees.
• Existing methods for inferring differentiation trajectories often assume tree-like structures and lack robust statistical approaches for determining the number of branching events.

Purpose of the Study:

• To develop a statistically sound method for identifying branching events in cell differentiation.
• To accurately determine the number and location of cell subgroups associated with branching and differentiation.

Main Methods:

• Introduction of K-Branches, a clustering algorithm that locally fits half-lines to single-cell data as proxies for differentiation branches.
• Application of a modified GAP statistic for model selection to determine the optimal number of lines describing the data.
• Evaluation on diverse datasets including single-cell RNA-Seq, single-cell qPCR, and artificial data.

Main Results:

• K-Branches successfully identifies the location and number of cell subgroups involved in branching events.
• The method demonstrates robust performance across various single-cell datasets, including myeloid progenitor differentiation, mouse blastocyst development, and human leukemia data.
• The algorithm provides a statistically principled way to model complex differentiation trajectories.

Conclusions:

• K-Branches offers an effective solution for inferring cell differentiation trajectories and identifying branching points.
• The method enhances the analysis of single-cell data by providing a data-driven approach to model developmental processes.
• An R implementation is available for broader scientific application.