Detecting disease comorbidity based on SNP association on PheWAS scale

Area of Science:

• Genetics
• Computational Biology
• Medical Informatics

Background:

• Disease comorbidity poses significant challenges in diagnosis and treatment.
• Understanding genetic factors, including single nucleotide polymorphisms (SNPs), is crucial for unraveling comorbidity.
• Existing methods require enhancement for accurate comorbidity prediction based on genetic data.

Purpose of the Study:

• To develop a machine learning method for detecting disease comorbidity using SNP association data.
• To improve comorbidity prediction accuracy by integrating disease-disease networks.
• To identify key SNPs contributing to comorbidity prediction.

Main Methods:

• Applied principal component analysis (PCA) for dimensionality reduction of high-dimensional SNP data.
• Developed a neural network model trained on reduced SNP vectors.
• Integrated a disease-disease network derived from SNP associations to compensate for information loss.
• Utilized a random forest classifier to identify informative SNPs.

Main Results:

• The neural network model trained on PCA-reduced SNP data outperformed the state-of-the-art method.
• Exploiting the disease-disease network significantly improved prediction performance in cross-validation.
• The SNP-based approach identified top contributing SNPs for comorbidity prediction.
• Identified statistically significant antagonistic and synergistic patterns in top SNPs.

Conclusions:

• The developed machine learning approach effectively detects disease comorbidity using SNP data.
• Integrating disease networks enhances predictive accuracy and aids in understanding comorbidity's genetic basis.
• The identified top SNPs offer insights into the root causes of comorbidity for specific disease pairs.