Order-restricted inference for ordered gene expression (ORIOGEN) data under heteroscedastic variances

Area of Science:

• Bioinformatics
• Statistical Genetics
• Genomics

Background:

• Gene expression analysis is crucial for understanding biological processes.
• Existing methods for time-course/dose-response data assume constant variability (homoscedasticity).
• Heteroscedasticity (varying variability) in gene expression data can impact analysis accuracy.

Purpose of the Study:

• To extend the order restricted inference (ORI) approach for gene expression data with heteroscedasticity.
• To develop a robust methodology for analyzing time-course and dose-response microarray data.
• To improve gene selection in the presence of varying expression variability.

Main Methods:

• An iterative algorithm was developed to estimate mean gene expression under order restrictions.
• A bootstrap-based methodology was employed for gene selection.
• The new method was compared to ORIOGEN using simulation studies.

Main Results:

• The proposed methodology effectively handles heteroscedasticity in gene expression data.
• Simulation studies confirmed the method maintains the false positive rate at the nominal level.
• The approach demonstrated competitive power for gene selection compared to ORIOGEN.

Conclusions:

• The extended ORI approach provides a robust framework for analyzing heteroscedastic gene expression data.
• This methodology enhances the accuracy of gene selection in time-course and dose-response studies.
• The approach is applicable to real-world biological data, as shown with breast cancer cell-line analysis.