Distributed transformer for high order epistasis detection in large-scale datasets.

Area of Science:

• Genetics
• Computational Biology
• Precision Medicine

Background:

• Understanding complex diseases requires analyzing genetic variations, including Single Nucleotide Polymorphisms (SNPs).
• Genome-Wide Association Studies (GWAS) identify SNPs linked to traits but often miss complex interactions (epistasis).
• Epistasis, or gene-gene interactions, is crucial for explaining most genetic diseases but computationally challenging to analyze.

Purpose of the Study:

• To develop a novel, explainable deep learning framework for detecting any-order epistasis in genomic data.
• To address the computational challenges and black-box nature of existing deep learning methods in genomic prediction.

Main Methods:

• A novel transformer-based framework for network interpretation was developed to analyze SNP combinations.
• The framework was designed for flexibility, portability, and scalability to handle large datasets.
• The approach was validated on three Wellcome Trust Case Control Consortium (WTCCC) datasets.

Main Results:

• The proposed framework demonstrated superior explainability compared to state-of-the-art methods.
• The method proved scalable to large datasets and portable across different deep learning accelerators.
• Key SNPs associated with known disease-related genes were successfully identified.

Conclusions:

• The novel transformer-based framework effectively detects and explains complex genetic interactions (epistasis).
• This approach advances precision medicine by providing interpretable insights into the genetic basis of complex diseases.
• The validated framework offers a scalable and portable solution for genomic prediction and epistasis analysis.