Question: Custom scoring for subnetworks in BioNet
0
3.4 years ago by
enricoferrero570
Switzerland
enricoferrero570 wrote:

BioNet seems to require p-values to generate its scores for each node:

library(BioNet)
library(DLBCL)
data(dataLym)

pvals <- cbind(t=dataLym$t.pval, s=dataLym$s.pval)
rownames(pvals) <- dataLym$label pval <- aggrPvals(pvals, order=2, plot=FALSE) fb <- fitBumModel(pval, plot=FALSE) scores <- scoreNodes(subnet, fb, fdr=0.001) module <- runFastHeinz(subnet, scores) I'm building a data integration pipeline where p-values and other factors will be integrated into a score that goes from -1 to +1. Can I use these custom scores directly as argument to runFastHeinz()? Or do I need to convert them to BioNet scores, somehow? The distribution of BioNet scores in this example seems to be much wider than [-1, 1]: summary(scores) Min. 1st Qu. Median Mean 3rd Qu. Max. -7.265 -6.946 -6.417 -5.610 -5.152 8.986 I'm also not entirely sure what the + and - sign mean in the BioNet scores: does the sign represents the directionality of the effect (I.e.: positive=activated, negative=inhibited)? Thank you. network bionet • 637 views ADD COMMENTlink modified 3.4 years ago by branislav misovic120 • written 3.4 years ago by enricoferrero570 Answer: Custom scoring for subnetworks in BioNet 1 3.4 years ago by Netherlands/Amsterdam branislav misovic120 wrote: hi again , so in your other post i mentioned the course i attended last year , check link there you can open this exercise: and in pdf they say " In input/Heinz Nodes FDR 7e-04.txt the nodes score that were used to generate the module referred to in the presentation are given. Observe that this time no p-values but actual scores are used. ...$ head Heinz_Nodes_FDR_7e-04.txt
#node score1
ENSMUSG00000057003 -5.07238096949341
ENSMUSG00000057000 -5.48958095921254
ENSMUSG00000002010 -4.16253938119165
....
To run heinz on input/nodes 7e-4.txt, we omit the parameters -a, -lambda and -FDR.This will take a few minutes.
"
So you can do it for sure ... (I also played with Fchange only  as that was my data)
If you open function scoreNodes  you will see call to

scoreFunction(fb, fdr=0.01)

...  and help for  fdr argument  says :
"... P-values below this threshold are considered to be significant and will score positively, p-values a bove the threshold are supposed to arise from the null model. The FDR can be used to control the size of the maximum scoring subnetwork,"

I just used some cutof and gave positive scores to my Fchanges and  bellow that i gave negative   ...

Maybe you can email the package  authors  the  link of your post and ask for more insight .

Best,

Branko