Data mining, neural nets, trees--problems 2 and 3 of Genetic Analysis Workshop 15.

Area of Science:

• Genetics
• Bioinformatics
• Computational Biology

Background:

• Genome-wide association studies (GWAS) and region-wide association studies (RWAS) are crucial for identifying disease susceptibility genes and predicting individual disease risk.
• The increasing importance of these tasks necessitates the examination of advanced data mining methods.

Purpose of the Study:

• To evaluate novel and existing data mining methods for classification and identification of disease susceptibility genes.
• To explore gene-by-gene and gene-by-environment interactions in genetic association studies.

Main Methods:

• Random forests were frequently employed due to their simplicity, robustness, and effectiveness in prediction and SNP screening.
• Logistic tree with unbiased selection was utilized as an alternative for efficient SNP selection.
• Ensemble machine learning methods were investigated for their potential in large-scale association studies.

Main Results:

• Random forests proved effective for prediction and initial SNP screening.
• Logistic trees offered an efficient alternative for selecting significant SNPs.
• Ensemble methods demonstrated potential for classification and handling complex interactions.

Conclusions:

• Machine learning, particularly ensemble methods, shows promise as a pre-screening tool for large-scale genetic association studies.
• These methods can mitigate overfitting, reduce computational time, and effectively incorporate higher-order interactions compared to traditional statistical approaches.
• Further development is needed for machine learning implementations to handle datasets with hundreds of thousands of SNPs.