Gene co-expression matrices (pairwise Pearson’s correlation, r, across Braak stage-related regions R1–R6) were calculated for each one of the six brain donors in the AHBA separately, and then combined into one consensus matrix based on the element-wise mean across all donors. Co-expression was converted to dissimilarity based on 1−r; in this way only positively co-expressed genes were taken into account. All genes were hierarchically clustered using single, complete, and average linkage and co-expression modules were obtained with the cutreeDynamicTree-function in the dynamicTreeCut R-package 1.63; minimum module size was set to 50 by default. The weighted correlation network analysis (WGCNA) R-package version 1.64.146 was used for further analysis of the modules. Hierarchical clustering by average linkage resulted in an acceptable number of missing genes while retaining the maximum number of modules (Supplementary Fig. 15; 167 modules with sizes up to 297 genes). For each module, the eigengene was obtained based on the first principle component and thus summarizes the expression of all genes within a module across all samples in Braak stage-related regions R1–R6. This was done for each brain donor
