In isglobal-brge/dsOmicsClient: DataSHIELD client site Omics association functions.

The same analysis can be performed for all features (e.g. CpGs) just avoiding the feature argument. This process can be parallelized using mclapply function from the multicore package.

ans <- ds.lmFeature(model = ~ diagnosis + Sex, 
                    Set = "methy",
                    datasources = conns,
                    mc.cores = 20)

This method corresponds to the pooled analysis approach and can be very time consiming since the function repeatedly calls the DataSHIELD function ds.glm(). We can adopt another strategy that is to run a glm of each feature independently at each study using limma package (which is really fast) and then combine the results (i.e. meta-analysis approach).

ans.limma <- ds.limma(model = ~ diagnosis + Sex,
                      Set = "methy", 
                      datasources = conns)

Then, we can visualize the top genes at each study (i.e server) by

lapply(ans.limma, head)

The annotation can be added by using the argument annotCols. It should be a vector with the columns of the annotation available in the ExpressionSet or RangedSummarizedExperiment that want to be showed. The columns of the annotation can be obtained by

ds.fvarLabels("methy")

Then we can run the analysis and obtain the output with the chromosome and gene symbol by:

ds.rm("ans.limma")

ans.limma.annot <- ds.limma(model = ~ diagnosis + Sex,
                            Set = "methy", 
                            annotCols = c("CHR", "UCSC_RefGene_Name"),
                            datasources = conns)

lapply(ans.limma.annot, head)

Then, the last step is to meta-analyze the results. Different methods can be used to this end. We have implemented a method that meta-analyze the p-pvalues of each study as follows:

ans.meta <- metaPvalues(ans.limma)
ans.meta

We can verify that the results are pretty similar to those obtained using pooled analyses. Here we compute the association for two of the top-CpGs:

res1 <- ds.lmFeature(feature = "cg13138089", 
                     model = ~ diagnosis + Sex, 
                     Set = "methy",
                     datasources = conns)
res1

res2 <- ds.lmFeature(feature = "cg13772815", 
                     model = ~ diagnosis + Sex, 
                     Set = "methy",
                     datasources = conns)
res2

We can create a QQ-plot by using the function qqplot available in our package.

qqplot(ans.meta$p.meta)

Here In some cases inflation can be observed, so that, correction for cell-type or surrogate variables must be performed. We describe how we can do that in the next two sections.

isglobal-brge/dsOmicsClient documentation built on March 20, 2023, 3:52 p.m.