Rough Sets for Phenotype-Based Prioritization of Causative Variants

Genome-wide association studies (GWAS) are essential for understanding the genetic basis of complex traits by identifying single nucleotide polymorphisms (SNPs) associated with phenotypes of interest. GWAS employ statistical methods to identify SNPs associated with phenotypes above a predetermined significance threshold. However, this threshold approach may inadvertently exclude highly significant SNPs, posing a potential limitation. GWAS datasets contain many SNPs, which can lead to ambiguity in association results. To resolve these discrepancies, several feature selection (FS) methods have been implemented prior to association tests. However, these FS methods do not effectively illustrate significant biological relevance of the resulting SNPs. Our work introduces a pipeline that combines a feature selection strategy based on the Rough Set theory with an association test using a machine learning approach. This innovative approach is applied to identify SNPs associated with blood cholesterol levels, focusing on low-density and high-density lipoprotein (LDL and HDL) cholesterol. The efficiency of the pipeline is evaluated using a cohort dataset from the American population to showcase the comparative efficacy of the pipeline. Our pipeline demonstrates excellent performance on datasets with low sample sizes, outperforming existing PLINK approach. Moreover, to enhance the biological relevance of selected SNPs, we extend our investigation to closely related SNPs, followed by rigorous enrichment studies annotating genes, biological processes, and pathways. This comprehensive exploration unveils the intricate cellular mechanisms and genetic determinants influencing LDL and HDL cholesterol levels. Our findings not only contribute valuable insights to the understanding of these traits but also suggest potential personalised treatment modalities.