A co-expression network based molecular characterization of genes responsive for Braak stages in Parkinson's disease.

The progression of Parkinson's disease (PD) is defined by six Braak stages. We used transcriptome data from PD patients with Braak stage information to understand underlying molecular mechanisms for the progress of the disease. We created networks of genes with continuously decreased/increased co-expression from control group to Braak 5-6 stages. These networks are significantly associated with PD-related mechanisms such as mitochondrial dysfunction and synaptic signalling among others. Applying weighted gene co-expression network analysis (WGCNA) algorithm to the co-expression networks led to more specific modules enriched with neurodegeneration-related disease pathways, seizure, abnormality of coordination, and hypotonia. Furthermore, we showed that one of the co-expression networks is clustered into three major communities with dedicated molecular functions: (i) tubulin folding pathway, gap junction-related mechanisms, neuronal system; (ii) synaptic vesicle, intracellular vesicle, proteasome complex, PD genes; (iii) energy metabolism, mitochondrial mechanisms, oxidative phosphorylation, TCA cycle, PD genes. The co-expression relations we identified in this study as crucial players in the disease progression cover several known PD-associated genes and genes whose products are known to physically interact with alpha-synuclein protein.

Effect of RNA-Seq data normalization on protein interactome mapping for Alzheimer's disease.

High throughput RNA sequencing brings new perspective to the elucidation of molecular mechanisms of diseases. Normalization is the first and most important step for RNA-Seq data, and it can differ based on the purpose of the analysis. Within-sample normalization methods (eg. TPM) are preferred when genes in a sample are compared with each other, and between-sample normalization methods (eg. deseq2, TMM, Voom) are used when the samples in a dataset are compared. Normalization approaches rescale the data, and, therefore, they affect the results of the analysis. Here, we selected two most commonly used Alzheimer's disease RNA-Seq datasets from ROSMAP and Mayo Clinic cohorts and mapped the differentially expressed genes on human protein interactome to discover disease-specific subnetworks. To this end, the raw count data were first processed with four different, commonly used RNA-Seq normalization methods (deseq2, TMM, Voom and TPM). Then, covariate adjustment was applied to the normalized data for gender, age of death and post-mortem interval. Each normalized dataset was separately mapped on the human protein-protein interaction network either in covariate-adjusted or non-adjusted form. Capturing known Alzheimer's disease genes and genes associated with the disease-related functional terms in the discovered subnetworks were the criteria to compare different normalization methods. Based on our results, applying covariate adjustment has a positive effect on normalization by removing the confounder effects. Covariate-adjusted TMM and covariate-adjusted deseq2 methods performed better in both transcriptome datasets.

Personalized Protein-Protein Interaction Networks Towards Unraveling the Molecular Mechanisms of Alzheimer's Disease.

Alzheimer's disease (AD) is a highly heterogenous neurodegenerative disease, and several omic-based datasets were generated in the last decade from the patients with the disease. However, the vast majority of studies evaluate these datasets in bulk by considering all the patients as a single group, which obscures the molecular differences resulting from the heterogeneous nature of the disease. In this study, we adopted a personalized approach and analyzed the transcriptome data from 403 patients individually by mapping the data on a human protein-protein interaction network. Patient-specific subnetworks were discovered and analyzed in terms of the genes in the subnetworks, enriched functional terms, and known AD genes. We identified several affected pathways that could not be captured by the bulk comparison. We also showed that our personalized findings point to patterns of alterations consistent with the recently suggested AD subtypes.