Bayesian inference with historical data-based informative priors improves detection of differentially expressed genes

Area of Science:

• Genomics
• Statistical Inference
• Bioinformatics

Background:

• High-throughput biotechnologies like microarrays generate vast amounts of data.
• The common 'large p, small n' problem arises from limited samples in high-throughput experiments.
• Existing data repositories offer potential for improving new data analyses.

Purpose of the Study:

• To investigate the feasibility and effectiveness of using historical data to derive informative priors.
• To apply this strategy to the problem of detecting differentially expressed genes in microarray data.
• To offer a practical solution for the 'large p, small n' challenge in genomics.

Main Methods:

• Utilizing a Bayesian framework to incorporate historical data into informative priors.
• Developing and applying a novel strategy for deriving priors from existing genomics datasets.
• Validating the approach through simulations and real-world microarray data analysis.

Main Results:

• The proposed strategy significantly outperforms existing methods, including state-of-the-art Bayesian hierarchical models.
• Demonstrated effectiveness in detecting differentially expressed genes.
• Successful exploitation of large-scale genomics data for statistical inference.

Conclusions:

• Deriving informative priors from historical data is a feasible and effective strategy.
• The method provides a promising practical approach to address the 'large p, small n' problem.
• Highlights the benefits of utilizing accumulated genomics big data for enhanced statistical analysis.