Using machine learning to identify proteomic and metabolomic signatures of stroke in atrial fibrillation.

Atrial fibrillation (AF) is a common cardiac arrhythmia, with stroke being its most detrimental comorbidity. The exact mechanism of AF related stroke (AFS) still needs to be explored. In this study, we integrated proteomics and metabolomics platform to explore disordered plasma proteins and metabolites between AF patients and AFS patients. There were 22 up-regulated and 31 down-regulated differentially expressed proteins (DEPs) in AFS plasma samples. Moreover, 63 up-regulated and 51 down-regulated differentially expressed metabolites (DEMs) were discovered in AFS plasma samples. We integrated proteomics and metabolomics based on the topological interactions of DEPs and DEMs, which yielded revealed several related pathways such as arachidonic acid metabolism, serotonergic synapse, purine metabolism, tyrosine metabolism and steroid hormone biosynthesis. We then performed a machine learning model to identify potential biomarkers of stroke in AF. Finally, we selected 6 proteins and 6 metabolites as candidate biomarkers for predicting stroke in AF by random forest, the area under the curve being 0.976. In conclusion, this study provides new perspectives for understanding the progressive mechanisms of AF related stroke and discovering innovative biomarkers for determining the prognosis of stroke in AF.