Improving sensitivity of linear regression-based cell type-specific differential expression deconvolution with...

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Blood gene expression studies aim to find molecular differences between healthy and diseased individuals.
• Cellular composition variability in blood samples is a major challenge, often ignored, hindering accurate gene expression analysis.
• Existing methods use linear regression for cell type-specific differential expression but overlook data artifacts.

Purpose of the Study:

• To quantify the parameter space affecting the performance of linear regression for cell type-specific differential expression detection.
• To evaluate the impact of sample size, cell type proportion variability, and mean squared error on detection sensitivity.
• To develop a computational tool for improved cell type-specific differential expression analysis.

Main Methods:

• Evaluated the effect of sample sizes, cell type-specific proportion variability, and mean squared error on sensitivity using linear regression.
• Developed the R package LRCDE for gene-by-gene linear regression-based cell type-specific differential expression (deconvolution) detection.
• Computed per-gene t-statistics, p-values, sensitivity, mean squared error, and diagnostic metrics, accounting for variability in cell type-specific gene expression estimates.

Main Results:

• Each evaluated parameter (sample size, proportion variability, MSE) influenced the variability of cell type-specific expression estimates and detection sensitivity.
• The R package LRCDE was developed to perform deconvolution and assess differential expression on a per-gene basis.
• LRCDE computes gene-specific metrics including t-statistics, p-values, sensitivity, and mean squared error.

Conclusions:

• Linear regression-based cell type-specific differential expression detection sensitivity is gene-specific, depending on MSE, sample sizes, and target cell proportion variability.
• The LRCDE package, utilizing Welch's t-test, demonstrates higher sensitivity in detecting cell type-specific differential expression compared to a global False Discovery Rate (FDR) threshold.
• LRCDE effectively identifies differential expression at α < 0.05 that may be missed by conventional FDR < 0.3 cutoffs.