Modeling Persistent Trends in Distributions.

Area of Science:

• Computational Biology
• Statistical Genetics
• Single-Cell Genomics

Background:

• Single-cell RNA sequencing (scRNA-seq) generates high-dimensional data capturing cell-to-cell variability.
• Analyzing temporal scRNA-seq data requires distinguishing true biological progression from technical noise.
• Existing methods often overlook distributional properties or sequential ordering inherent in time-course data.

Purpose of the Study:

• To develop a nonparametric framework for modeling sequences of probability distributions in time-course scRNA-seq data.
• To differentiate between sequential progression effects and confounding noise.
• To estimate gene expression trends indicative of biological process progression.

Main Methods:

• Introduced a novel regression model for ordinal covariates with univariate distributions as responses.
• Formalized 'trend' in distributions as linearity under the Wasserstein metric.
• Implemented the framework using a fast alternating projections algorithm.

Main Results:

• The proposed method effectively distinguishes persistent trends from random variations in simulated data.
• Demonstrated the framework's utility in analyzing real single-cell gene expression data.
• Successfully identified genes associated with biological process progression.

Conclusions:

• The nonparametric framework provides a robust approach for analyzing temporal single-cell gene expression data.
• The method accurately models distributional changes over time, capturing underlying biological trends.
• This approach enhances the ability to discover genes driving biological processes in dynamic cellular systems.