Statistical analysis of correlated expression data from high throughput experiments.

Area of Science:

• Genomics
• Bioinformatics
• Statistical Genetics

Background:

• High throughput experimental data often contain complex feature dependencies.
• Ignoring these dependencies can lead to inflated false discovery rates (FDR), reduced statistical power, and biased interpretations.
• Accurate detection of biological signals necessitates proper accounting for feature dependencies.

Purpose of the Study:

• To introduce a novel statistical method, Analysis of Correlated Expressions (ACE), for comparing mean expression of features between two groups.
• To address the challenges posed by feature dependencies and variance heterogeneity in high throughput data.
• To provide a scalable and parameter-free method for robust biological signal detection.

Main Methods:

• ACE employs a factor analytic model to capture feature dependencies.
• The method incorporates heterogeneity of variances between groups.
• ACE does not assume normal distribution of the data and is scalable.

Main Results:

• Extensive simulations show ACE is more powerful than existing methods while controlling FDR.
• ACE successfully identified novel findings in microRNA, neuroblastoma gene expression, and Huntington's disease datasets.
• The method demonstrated superior performance in controlling FDR and increasing statistical power.

Conclusions:

• ACE offers a powerful and robust approach for analyzing high throughput data with complex dependencies.
• The method enhances the accuracy of biological signal detection and interpretation.
• ACE provides a valuable tool for researchers in genomics and bioinformatics.