Airpart: interpretable statistical models for analyzing allelic imbalance in single-cell datasets

Area of Science:

• Genomics
• Computational Biology
• Statistical Genetics

Background:

• Allelic expression analysis reveals cis-regulatory mechanisms driving genetic variation.
• Measuring allelic imbalance (AI) in bulk data can obscure cell-type-specific (CTS), spatial, or temporal AI signals due to lack of resolution.

Purpose of the Study:

• To introduce a statistical method, airpart, for identifying differential CTS AI from single-cell RNA-sequencing (scRNA-seq) data.
• To enable the detection of dynamic AI patterns in spatially or time-resolved datasets.

Main Methods:

• airpart employs a Generalized Fused Lasso with Binomial likelihood to partition cells by AI signal, accounting for low counts in scRNA-seq data.
• A hierarchical Bayesian model is utilized for AI statistical inference.
• The method outputs discrete data partitions, highlighting groups of genes and cells under common cis-genetic regulatory mechanisms.

Main Results:

• Simulations demonstrated airpart's accuracy in detecting cell type partitions by AI and its lower Root Mean Square Error (RMSE) for allelic ratio estimates compared to existing methods.
• Real data analysis revealed differential AI patterns across cell states.
• airpart successfully defined trends of AI signal over spatial or temporal axes.

Conclusions:

• airpart provides a robust statistical framework for analyzing AI in high-resolution datasets.
• The method facilitates the discovery of cis-regulatory mechanisms by dissecting cell-type-specific and dynamic AI.
• airpart is available as an R/Bioconductor package.