Single Nucleotide Polymorphism relevance learning with Random Forests for Type 2 diabetes risk prediction.

Area of Science:

• Genetics
• Medical Informatics
• Computational Biology

Background:

• Single Nucleotide Polymorphisms (SNPs) are crucial in disease development.
• Artificial intelligence (AI) offers potential for early disease diagnosis and prevention.
• Identifying disease-associated SNPs can aid clinical decision-making.

Purpose of the Study:

• To identify SNPs associated with Type 2 diabetes using AI.
• To develop a decision-support tool for predicting Type 2 diabetes risk.
• To compare the performance of different machine learning techniques for SNP analysis.

Main Methods:

• Random Forest (RF) was employed to determine the importance of SNPs.
• Support Vector Machines (SVM) and Logistic Regression were used for comparison.
• k-Nearest Neighbour (k-NN) was utilized for prediction, weighting attributes by RF relevance.

Main Results:

• RF achieved an area under the ROC curve of up to 0.89 for risk prediction.
• RF demonstrated superior prediction accuracy and attribute relevance stability compared to SVM and Logistic Regression.
• The study analyzed a dataset of 677 subjects.

Conclusions:

• Random Forest is an effective method for identifying relevant SNPs without making prior assumptions.
• AI-driven SNP analysis can enhance predictive models for diseases like Type 2 diabetes.
• The findings support the use of RF in personalized medicine and risk assessment.