Meta-Analyzing Multiple Omics Data With Robust Variable Selection.

Area of Science:

• Genomics
• Bioinformatics
• Statistical Genetics

Background:

• High-throughput omics data are increasingly prevalent across scientific disciplines.
• Meta-analysis synthesizes multiple datasets for robust model estimation and prediction.
• Variable selection is crucial for high-dimensional omics meta-analysis but existing methods struggle with outliers.

Purpose of the Study:

• To develop a robust variable selection algorithm for meta-analyzing high-dimensional omics datasets.
• To address the sensitivity of existing methods to outliers and missed covariate detections.
• To enhance the reliability of meta-analysis results in the presence of data anomalies.

Main Methods:

• A novel robust variable selection algorithm based on logistic regression for meta-analysis.
• Identification of an outlier-free subset from each dataset using least trimmed squared estimates and hierarchical bi-level selection.
• Refinement using a reweighting step to improve efficiency on the identified non-outlier subset.

Main Results:

• The proposed method demonstrates superior performance in simulation studies and real data analysis.
• The algorithm provides more reliable results than existing meta-analysis methods when outliers are present.
• Effective handling of outliers leads to improved accuracy in variable selection for omics data.

Conclusions:

• The developed robust variable selection algorithm enhances the reliability of high-dimensional omics meta-analysis.
• This approach offers a significant improvement over existing methods, particularly in datasets with outliers.
• The method facilitates more accurate insights from synthesized omics data, advancing biological and medical research.