Deep-Neural-Network-Aided Genetic Association Testing in Samples with Related Individuals.

Area of Science:

• Genetics
• Bioinformatics
• Machine Learning

Background:

• Genome-wide association studies (GWAS) are crucial for identifying genetic loci linked to complex traits and diseases.
• Machine learning (ML) can expand GWAS analytical capabilities, but deep learning (DL) is underutilized, especially with cryptic relatedness.
• Traditional GWAS methods may face challenges with complex genetic architectures and relatedness.

Purpose of the Study:

• To propose a novel deep neural network (DNN)-based machine learning method for genetic association testing in samples with related individuals.
• To enhance the identification of genetic variants associated with complex traits by improving predictive performance.
• To complement traditional GWAS frameworks and increase power in detecting genetic associations.

Main Methods:

• Developed a DNN-based ML method to approximate phenotype-genotype relationships in association tests.
• Combined approximations from multiple tests to improve variant identification.
• Validated the method through simulation studies and application to the Framingham Heart Study dataset.

Main Results:

• The DNN-based method effectively complements conventional statistical approaches in GWAS.
• The proposed method generally achieves increased statistical power for detecting genetic associations.
• Applied to Framingham Heart Study data, the method identified genome-wide SNPs associated with average systolic blood pressure.

Conclusions:

• Deep learning offers a powerful augmentation for traditional GWAS, particularly in complex sample structures.
• The proposed DNN method enhances the ability to detect genetic associations in the presence of relatedness.
• This approach facilitates the discovery of genetic variants influencing complex traits, such as systolic blood pressure.