Robust kernel canonical correlation analysis to detect gene-gene co-associations: A case study in genetics

Area of Science:

• Genetics
• Bioinformatics
• Statistical analysis

Background:

• Kernel canonical correlation analysis based U-statistic (KCCU) detects nonlinear gene-gene co-associations.
• Estimating KCCU variance is computationally intensive, and kernel CCA is not robust to data contamination.
• Robust kernel methods offer a potential solution for analyzing noisy genetic data.

Purpose of the Study:

• To develop a robust non-parametric KCCU for detecting gene-gene co-associations in contaminated datasets.
• To propose an influence function-based estimator for KCCU variance.
• To demonstrate the superior performance of robust kernel CCA in identifying gene interactions, particularly for complex diseases like schizophrenia.

Main Methods:

• Development of a robust kernel mean element and a robust kernel (cross)-covariance operator.
• Implementation of a non-parametric robust KCCU less sensitive to noise.
• Application of influence function-based estimation for KCCU variance.
• Validation using synthesized and real-world data from the Mind Clinical Imaging Consortium (MCIC).

Main Results:

• The proposed robust KCCU method shows increased statistical power with larger sample sizes.
• Robust test statistics provide incremental power gains compared to standard KCCU.
• Analysis of MCIC schizophrenia data identified 768 candidate genes and significant gene pairs, revealing potential gene-gene interaction networks.
• The robust methods successfully identified previously undiscovered genes and outperformed existing approaches.

Conclusions:

• The developed robust kernel CCA offers a computationally efficient and noise-tolerant approach for detecting gene-gene co-associations.
• This method enhances the identification of gene-gene interactions crucial for understanding complex diseases like schizophrenia.
• The robust approach demonstrates superior performance and potential for discovering novel genetic insights over current methods.