Ge-SAND: an explainable deep learning-driven framework for disease risk prediction by uncovering complex genetic...

Area of Science:

• Genomics
• Computational Biology
• Precision Medicine

Background:

• Accurate genetic risk prediction is crucial for effective disease intervention and personalized medicine.
• Current methods struggle to capture complex genetic interactions, especially with large datasets.
• The curse of dimensionality limits the analysis of extensive pairwise genetic combinations.

Purpose of the Study:

• To develop an explainable deep learning framework for uncovering complex genetic interactions.
• To improve disease risk prediction by analyzing large-scale genotype-phenotype relationships.
• To provide insights into pathogenic mechanisms underlying complex diseases.

Main Methods:

• Developed Ge-SAND (Genomic Embedding Self-Attention Neurodynamic Decoder), a deep learning framework.
• Utilized genotype and genomic positional information to identify intra- and interchromosomal interactions.
• Applied the framework to simulated data and UK Biobank cohorts for Crohn's disease, schizophrenia, and Alzheimer's disease.

Main Results:

• Ge-SAND achieved up to a 20% improvement in AUC-ROC compared to mainstream methods.
• Identified large-scale genotype relationships and revealed genetic mechanisms through self-attention networks.
• Discovered novel genetic interaction pairs, such as ISOC1 and HOMER2, suggesting a brain-gut axis link in certain diseases.

Conclusions:

• Ge-SAND is a novel deep learning approach for capturing large-scale genetic interactions.
• The framework integrates disease risk prediction with interpretable genetic mechanism insights.
• Ge-SAND advances genomic research and precision medicine by offering a valuable analytical tool.