Statistical Inference of Cell-type Proportions Estimated from Bulk Expression Data.

Area of Science:

• Genomics
• Bioinformatics
• Statistical Modeling

Background:

• Cell-type-specific analysis of bulk samples is increasingly important.
• Accurate estimation of cell-type proportions is a critical prerequisite.
• Quantifying uncertainty in these estimations remains understudied, potentially impacting downstream analyses.

Purpose of the Study:

• To introduce a statistical deconvolution framework for estimating cell-type proportions.
• To develop a method that quantifies the uncertainties associated with these estimations.
• To improve the accuracy of cell-type-specific differential gene expression analysis.

Main Methods:

• Developed a flexible statistical deconvolution framework.
• Proposed a decorrelated constrained least squares method (DECALS).
• DECALS estimates cell-type proportions and their sampling distribution, allowing for subject-specific covariance.

Main Results:

• Simulation studies showed DECALS accurately quantifies estimation uncertainties, unlike other methods.
• Applied DECALS to post-mortem brain samples (ROSMAP, GTEx).
• Accounting for uncertainty improved identification of cell-type-specific differentially expressed genes/transcripts in disease and sex comparisons.

Conclusions:

• DECALS provides a robust method for estimating cell-type proportions and their uncertainties.
• Considering uncertainty enhances the accuracy of cell-type-specific differential expression analysis.
• This approach is valuable for analyzing complex biological samples like brain tissue.