D iii) the ten data sets combined. Each P-value was log transformed (base 10) and

D iii) the ten data sets combined. Each P-value was log transformed (base 10) and multiplied by minus the sign of the correlation coefficients. For example, logPvalueWB takes on a large positive (negative) number for CpG probes that have a significant positive (negative) correlation with age across the six WB data sets. Analogously, logPvalueBrain and logPvalueAll measure age associations in the brain data sets and in all ten data sets, respectively. Additional file 1 shows a scatterplot involving correlation test P-values for age effects in schizophrenia cases and healthy controls based on the Dutch WB data sets (data sets 2 and 3). Note that meta analysis P-values for schizophrenics (cases) are highly correlated (r = 0.78) with those of healthy controls (y-axis). Thus, Additional file 1 shows that schizophrenia disease status has a negligible effect on aging-related changes for the vast majority of CpG sites. Additional file 2 shows scatterplots of correlation test P-values for measuring aging effects on DNA methylation profiles in the different brain regions (DNA methylation data sets 7 to 10). Overall, these P-values are highly correlated, which shows that age has a similar effect in all four brain regions. Having said this, comparisons involving the cerebellum (labeled CRBLM) show weaker correlations. Future studies involving additional cerebellum samples could address whether these systematic aging differences reflect the histologically distinct composition of the cerebellum or rather reflect sample quality issues such as degradation of DNA.Consensus module analysis with WGCNAAlthough many platforms exist for measuring methylation levels [22], the 16 DNA methylation data sets considered here were measured on the Illumina platform (Table 1). Data sets 1 through 10 were used in a consensus network analysis while the remaining data sets were used for validation. We analyzed 4 novel blood data sets (labeled 1 to 3, and 11) and 12 additional public data sets. While most of the first ten data sets used in the consensus network PubMed ID:http://www.ncbi.nlm.nih.gov/pubmed/28859980 analysis involved healthy subjects, data set 3 involved blood tissue from schizophrenic cases. The effect of disease status on aging effects is discussed below and in Additional file 1. A more detailed description of the subject characteristics is provided in the Materials and methods section.Correlating CpG sites with age and standard meta analysisEach individual CpG marker on the array was correlated to age in each of the ten data sets. We used a robustWe used WGCNA to construct consensus modules across ten data sets (sets 1 to 10 in Table 1). Consensus modules group together methylation probes that are highly co-methylated across the ten input data sets (Materials and methods). Since consensus modules are, by definition, present in multiple independent data sets, they represent common (perhaps universal) and robust co-methylation relationships that reflect the SCH 530348 site underlying biology rather than technical artifacts. Weighted network methods are particularly useful for identifying consensus modules since they allow one to calibrate the individual networks. Further, they give rise to powerful module preservation statistics that can be used to determine whether modules can be validated in independent data sets [23,24]. Figure 3 shows the hierarchical cluster tree that results from consensus network analysis of sets 1 to 10. Branches in the tree correspond to consensus modules. The first color band underneath the tree i.