Entering edit mode

2.1 years ago

steve.booth
▴
40

Hi there,

We have 4 quantitative imaging measures and gene expression for the same tissues. So I am working on GSVA to see which pathways change in relation to changes in the pathology - based on the 4 imaging variables.

```
############# Create the GSVA obj #############
eDat_es <- gsva(eset, myC5, min.sz=10, max.sz=Inf, method="gsva", mx.diff=TRUE,verbose=FALSE)
adjPvalueCutoff <- 0.05
logFCcutoff <- log2(1.5)
############### Pathways associated with Alveolar surface area (SA) ###########
design <- model.matrix(~ SA + Age + Sex+ packyears +BMI, data = pDat)
fit <- lmFit(eDat_es, design)
fit <- eBayes(fit)
allGeneSets <- topTable(fit, coef=2, number=Inf)
SA_DEgeneSets <- topTable(fit, coef=2, number=Inf,
p.value=adjPvalueCutoff, adjust="BH")
resSA <- decideTests(fit, p.value=adjPvalueCutoff)
summary(resSA)
############### Pathways associated with Volume fraction of Alveoli (Alv) ###########
design <- model.matrix(~ Vv.alv.par. + Age + Sex+ packyears +BMI, data = pDat)
fit <- lmFit(eDat_es, design)
fit <- eBayes(fit)
allGeneSets <- topTable(fit, coef=2, number=Inf)
Alv_DEgeneSets <- topTable(fit, coef=2, number=Inf,
p.value=adjPvalueCutoff, adjust="BH")
resAlv <- decideTests(fit, p.value=adjPvalueCutoff)
summary(resAlv)
############### Pathways associated with Volume fraction of Septa (Septa) ###########
design <- model.matrix(~ Vv.tissue.par. + Age + Sex+ packyears +BMI, data = pDat)
fit <- lmFit(eDat_es, design)
fit <- eBayes(fit)
allGeneSets <- topTable(fit, coef=2, number=Inf)
Septa_DEgeneSets <- topTable(fit, coef=2, number=Inf,
p.value=adjPvalueCutoff, adjust="BH")
resSepta <- decideTests(fit, p.value=adjPvalueCutoff)
summary(resSepta)
############### Pathways associated with Volume fraction of Ducts (Duct) ###########
design <- model.matrix(~ Vv.duct.par. + Age + Sex+ packyears +BMI, data = pDat)
fit <- lmFit(eDat_es, design)
fit <- eBayes(fit)
allGeneSets <- topTable(fit, coef=2, number=Inf)
Duct_DEgeneSets <- topTable(fit, coef=2, number=Inf,
p.value=adjPvalueCutoff, adjust="BH")
resDuct <- decideTests(fit, p.value=adjPvalueCutoff)
summary(resDuct)
```

This results in 4 tables of enriched pathways (1x for each imaging measure): SA_DEgeneSets, Alv_DEgeneSets, Septa_DEgeneSets, Duct_DEgeneSets.

Is there a way that we could statistically test the overlap, or find a core set of enriched pathways? At the moment I just collapse all 4 tables on matching row names to find overlapping pathways....

Thanks in advance for any suggestions! Steve

Hey Steve, I guess that you could treat these signatures / pathways as 'genes' and just perform a meta-analysis with, e.g., RankProd. I have never seen this done before, so, I am only adding this as a comment. It is no problem to also do a manual overlap and report the findings.

As you are probably gathering, GSVA is a very powerful technique and there is much possibility with the data. One thing that I did recently was to input the signatures/pathways into a regression model and then pick the best pathway predictors of, e.g., response to treatment. The results made a lot of sense.

Hey Kevin, thanks for the reply. Seems that you are the on call guru for my posts on here at the moment.

I like the suggestion of the regression. One thing I have played with is trajectory inference analysis using SCORPIUS, to see if the DE Genes can predict a trajectory of disease severity among the samples. Also includes a random forest method to select the most important predictors. So perhaps I might try that with the pathways.

As you said, lots of possibilities with the data!