Year: 2026
Type: article
Abstract: Identification of genetic variations associated with complicated characteristics, including susceptibility to particular diseases, has become a crucial function of Genome-Wide Association Studies (GWAS). However, the challenge of predicting statistical Single Nucleotide Polymorphisms (SNPs) remains due to high-dimensional data and complex genetic architectures. In this study, seven regression models are assessed: Random Forest, Extra Trees Regressor, XGBoost, CatBoost, LightGBM, Support Vector Regressor, and Elastic Net for their predictive capacity and precision in predicting SNP significance using transformed P-values (−log10 scale) as a continuous regression target, with Bonferroni and False Discovery Rate corrections applied during preprocessing to define significance thresholds for interpretation. By transforming P-values using a logarithmic scale, we examine model performance based on various metrics, including mean squared error, mean absolute error, explained variance and R-squared, to determine which models best predict SNP significance. Research findings show that tree-based ensemble methods, particularly the Random Forest and Extra Trees Regressor, achieve the highest predictive accuracy, with Random Forest emerging as the top performer. Gradient boosting models, such as XGBoost and CatBoost, also demonstrate robust results, indicating their ability to capture complex SNP interactions. The research study provides insight into the selection of the model for GWAS-based predictions and contributes to methodologies for more accurately identifying SNPs with potential implications in disease risk prediction. The results obtained offer practical guidance for researchers in choosing appropriate regression models for high-dimensional genetic data.
Source: Premier journal of science.
DOI: 10.70389/pjs.100225
Language: English
Subfield: Genetics
Domain: Life Sciences
Open Access status: hybrid
OpenAlex ID: W7127620620