Super Exact test

To provide efficient computation of the statistical distributions of multi-omic pathway/module set intersections, we used the theoretical framework implemented in R package . A circle plot is returned with the frequency of all significant omic combinations and their significance levels.

mi <- runSupertest(moduleSummary, pvalueThr = 0.05, zscoreThr = 0.05, resampligThr = 80, 
    excludeColumns = c("pathway", "module", "resamplingCount"))

supertest plot

As you can see, the combination of expression and methylation is significant, this means that they co-regulate the same pathway modules more often than what expected by chance.

Frequency distribution with radial plot.

This plot shows the frequency distribution of pathways aggregated into macro categories, using Reactome or KEGG hierarchical structure, separately for each omic combinations. This plot suggests prognostic biological processes that may be impacted by the omics and their cross-talk.

omicsClasses2pathways <- computeOmicsIntersections(moduleSummary, pvalueThr = 0.05, 
    zscoreThr = 0.05, resampligThr = 80, excludeColumns = c("pathway", "module", 
        "resamplingCount"))

omicsClasses2pathways <- lapply(omicsClasses2pathways, stripModulesFromPathways)

omicsClasses2fathers <- lapply(omicsClasses2pathways, annotePathwayToFather, 
    graphiteDB = reactome, hierarchy = pathHierarchyGraph)

correspondence <- lapply(names(omicsClasses2pathways), function(omicClass) {
    data.frame(path = omicsClasses2pathways[[omicClass]], father = omicsClasses2fathers[[omicClass]], 
        stringsAsFactors = F)
})
MOMfreqDataframe <- houseOfClipUtility::computeFreqs(omicsClasses2fathers)

At this point, we have computed the frequencies. In the table of frequencies, we find pathways grouped together. As a fist example, we check where the combination of two or more omics points.

combiClass <- grep(";", MOMfreqDataframe$class)
MOMfreqDataframe.multi <- MOMfreqDataframe[combiClass, , drop = F]
houseOfClipUtility::plotFrequencies(MOMfreqDataframe.multi, minSize = 6, maxSize = 12, 
    width = 9, manualColors = c("#E5C002", "gray45", "#8C04B6", "#01B7BF"))

We confirmed that signaling pathways are mainstream mechanisms in our dataset, even at module level. But now, we can see that also cell cycle, extracellular matrix organization and metabolism are involved. We can also take a look at the single omics.

MOMfreqDataframe.single <- MOMfreqDataframe[-combiClass, , drop = F]
houseOfClipUtility::plotFrequencies(MOMfreqDataframe.single, manualColors = c(exp = "#E33402", 
    met = "#04B880", mut = "#3871CC"), minSize = 5, maxSize = 12, width = 9)

plot frequencies of pathways with only one omic

And we can look also to all the classes together.

houseOfClipUtility::plotFrequencies(MOMfreqDataframe)

plot frequencies of pathways with all omics

Pathway summary

We can plot a report of the first 10 pathways, sorted by pvalue of the Cox proportional hazard model.

plotMultiPathwayReport(multiOmicsFullHuge, 10)

The heatmap summarized the \(p-value\) of the Cox proportional hazard model using all the covariates. The first column is the \(p-value\) of the model, the others columns contain the \(p-value\) of the single covariates. Colorscale is used to graphically visulize the \(p-values\). The \(p-value\) heatmap helps to figure out the involvement of different omics.

Pathway Heatmap

We can explore also the most important genes of the pathway and their original pattens using an heatmap plot of the original values. The most important genes are selected as described in Martini et al.

plotPathwayHeat(multiOmicsFullHuge[["VEGFA-VEGFR2 Pathway"]], sortBy = c("expPC2", 
    "met3k"), paletteNames = c(exp = "red", met = "green", mut = "blue"), nrowsHeatmaps = 2, 
    withSampleNames = F, fontsize_row = 6)

## 'select()' returned 1:1 mapping between keys and columns
## 'select()' returned 1:1 mapping between keys and columns

Pathway Kaplan Meier

Then we can also check differences in survival using Kaplan Meier curves dividing patients in groups with different omics patterns (e.g. patients with different methylation pattern and high and low levels of PC2 in expression).

plotPathwayKM(multiOmicsFullHuge[["VEGFA-VEGFR2 Pathway"]], formula = "Surv(days, status) ~ met3k + expPC2", 
    paletteNames = "Paired")

Plot Pathway Summary of resampling results

Alternatively, we can visualize the table as an heatmap of \(p-value\). In this case, we plot only those pathways with a resampling success score greater than 70 (or equal to 70).

resamplingTrue <- names(which(resampligSuccessCount >= 70))
plotMultiPathwayReport(multiOmicsFullHuge[resamplingTrue], 39, fontsize = 10, 
    fontsize_number = 0, fontsize_row = 8, fontsize_col = 8)

Super test

In second instance, we can ask if two or more omics are significant in the same pathway simultaneously and if this omic interaction is more frequent than those expected by chance. To perform this test we use the runSupertest (SuperExactTest R package).

moInt <- MOSClip::runSupertest(pathwaySummaryHuge, pvalueThr = 0.05, zscoreThr = 0.05, 
    resampligThr = 80, excludeColumns = c("resamplingCount"))

supertest plot

As you can see from the -log10 pvalues, we found that there are 6 pathways shearing expression and methylation omics that are both significant as coefficients.

Frequency plot

As, we did for modules we can group pathway by hierarchy and representing the contribution of different omics to survival prediction in different biological processes.

omicsClasses2pathways <- computeOmicsIntersections(pathwaySummaryHuge, resampligThr = 70, 
    excludeColumns = c("resamplingCount"))
omicsClasses2fathers <- lapply(omicsClasses2pathways, annotePathwayToFather, 
    graphiteDB = reactome, hierarchy = pathHierarchyGraph)
freqDataframe <- houseOfClipUtility::computeFreqs(omicsClasses2fathers)

Now we have a table of frequencies, class and category to use to have some hits of the combinations of the differents omics tested. As a first example, we check where the combination of two or more omics points.

plotFrequencies(freqDataframe, minSize = 6, maxSize = 12, width = 10, manualColors = c("#E33402", 
    "#E5C002", "gray45", "#8C04B6", "#04B880", "#01B7BF", "#3871CC"))

We have discovered that our pathway analysis tells us that “Cell cycle” is the most involved category for mutation and methylation.

Kaplan Meier curves

Maybe now we are interested to see the interaction in relation to the survival. We peak the pathway “FGFR1 mutant receptor activation” and we plot the Kaplan Meier curves exploriong the interaction between mutation profile and expression (PC2).

plotPathwayKM(multiOmicsFullHuge[["FGFR1 mutant receptor activation"]], formula = "Surv(days, status) ~ mut + expPC2", 
    paletteNames = "Paired", risk.table = T, size = 2, inYears = T)

Kaplain-Meier of FGFR1 mutant receptor activation

Predictive genes heatmap

Finally, we can plot the heatmap of the most predictive gene in the pathway. When looking to predictive gene it is extremely useful to correlate survival behavior (death event or survival time) to predictive genes as well as other annotation such as tumor grade. That’s why, we allows the user to plot heatmap with additional custom annotations. The data frame that we produced can be used in the heatmap of the predictive genes.

plotPathwayHeat(multiOmicsFullHuge[["FGFR1 mutant receptor activation"]], sortBy = c("expPC2", 
    "mut", "PFS"), paletteNames = c(exp = "red", met = "green", mut = "blue"), 
    additionalAnnotations = additionalA, additionalPaletteNames = list(PFS = "yellow", 
        years = "teal"), nrowsHeatmaps = 2, withSampleNames = F)

## 'select()' returned 1:1 mapping between keys and columns
## 'select()' returned 1:1 mapping between keys and columns

Heatmap of Negative regulation of the PI3K/AKT network