Combining supervised and unsupervised machine learning with Mendelian randomization to predict phase separation-related biomarkers associated with renal failure

Abstract Background Although Genome-Wide Association Studies (GWAS) link genetic variants to kidney disease, the specific role of phase separation-related genes (PSGs) in renal failure pathogenesis is unknown. This study investigates the potential of PSGs as novel biomarkers to improve the early detection and treatment of renal failure. Methods We created ten gene co-expression modules and associated cluster trees using enrichment and differential expression analysis. The final phase separation-related biomarkers were screened using three machine classification algorithms, and each gene's functional pathways and relationship to immune function were examined. In order to thoroughly confirm the link between potentially linked genes and the onset of renal failure, the study also employed Mendelian Randomization (MR) and supervised and unsupervised machine learning. It also showed the statistical power of unsupervised learning and the results of additional verification. Results According to the correlation of gene expression, ten genes that may be related to renal failure were screened out. Linking phase separation to renal failure identified two core phase separation-related genes, ARL6IP4 and MRRF. Mendelian randomization provided suggestive evidence of a potential causal association between genetically predicted constipation and increased risk of renal failure. Conclusions This study points to a higher potential of MRRF as a biomarker for renal failure than ARL6IP4. There may also be a potential causal association between the prevalence of constipation and the incidence of renal failure.