Bayesian screening for feature selection.

Area of Science:

• Biostatistics
• Genomics
• Bioinformatics

Background:

• Genome-wide association studies (GWAS) analyze high-dimensional data to find significant features.
• Identifying rare, weakly expressed, and continuous-valued features is crucial for detailed analysis in biomedical research.

Purpose of the Study:

• To develop an exact, rapid Bayesian screening approach for identifying important features in large biomedical datasets.
• To focus on the missed discovery rate (MDR) rather than the false discovery rate (FDR) for improved feature identification.

Main Methods:

• Utilized a Gaussian random mixture model for a Bayesian screening approach.
• Focused on calculating the missed discovery rate to assess the probability of failing to identify informative features.
• Incorporated clinical and regulatory priorities to determine critical values.

Main Results:

• The proposed method offers an exact and rapid Bayesian screening with attractive diagnostic properties.
• It provides the likelihood of a feature meriting further investigation.
• The approach assesses effect magnitudes and the proportion of features with similar expected responses.

Conclusions:

• The Bayesian screening method effectively identifies important features in high-dimensional biomedical data.
• Focusing on the missed discovery rate offers a valuable alternative to traditional false discovery rate control.
• The method's diagnostic properties and adaptability to clinical priorities enhance its utility in genetic association studies.