Monte Carlo cross-validation analysis screens pathway cross-talk associated with Parkinson's disease

Abstract

We purposed to identify underlying functional pathway cross-talk in Parkinson's disease (PD) through Monte Carlo cross-validation analysis. Microarray data set of E-GEOD-6613 was downloaded from ArrayExpress database. First, the identification of differentially expressed genes (DEGs) was implemented, following by extracting the potential disrupted pathway enriched by DEGs. In addition, a discriminating score (DS) was computed based on the distribution of gene expression levels by quantifying their pathway cross-talk for each pair of pathways. Furthermore, random forest (RF) classification model was utilized to identify the top ten paired pathways with high AUC between PD and healthy control samples using the tenfold cross-validation method. Finally, Monte Carlo cross-validation was repeated 50 times to explore the best pairs of pathways. After quantile normalization, a total of 9331 genes with higher than 0.25-fold quantile average across all samples were obtained. Totally, 42 DEGs and 19 differential pathways enriched from DEGs were identified. We then ranked each pathway according to their AUC values, the pair of pathways, phosphatidylcholine biosynthesis I, and PPAR signaling obtained the best AUC value of 0.942. Moreover, the paired pathways of mTOR signaling and CD28 signaling in T helper cells had higher AUC value of 0.837 in five bootstraps. Two paired pathways, including phosphatidylcholine biosynthesis I and PPAR signaling, as well as mTOR signaling and CD28 signaling in T helper cells were able to accurately classify PD and healthy control samples. Significantly, these paired pathways might be underlying biomarkers for early diagnosis and therapy of PD.

Keywords: Differentially expressed genes; Monte Carlo cross-validation; Parkinson’s disease; Pathway cross-talk; Random forest classification.