up regulated and down regulated genes in co-expression network

Question

[au.rinki.bio]

hi all, I am constructing a co-expression network using gene microarray data via Pearson correlation. I consider a threshold value of absolute 0.80 for the correlation coefficient. I considered three microarray datasets create co-expression network for each dataset separately after that I merged the network in order to find out common node and link shared by three datasets. after that, I found that the merged network has two components with maximum interaction when we screened the co-expression network, we saw that all the upregulated genes are in one component and all the dow regulated another component then my question is that is that possible?

[Kevin Blighe]

Possible, but not probable. I have used absolute 0.8 previously, and never encountered this issue.

Please show the code that you used to create the network.

Also, what is your input data's distribution?

[au.rinki.bio]

This is the which i had used for co-exoression network construction

 library("psych")
    data1<-read.table("gse7803_comp.txt")
    head(data1)
    Norm.interest.corr1 <- corr.test( t(data1), method="pearson", ci=F)
    Norm.interest.corr
    names(Norm.interest.corr1)
    Norm.interest.corr1$p[lower.tri( Norm.interest.corr1$p,diag=TRUE)]=NA
    Pval.adj <- as.data.frame(as.table(Norm.interest.corr1$p))
    Norm.interest.corr1$r [lower.tri( Norm.interest.corr1$r,diag=TRUE)]=NA
    Correlation <- as.data.frame(as.table(Norm.interest.corr1$r))
    Cor.table <- na.exclude(cbind( Correlation, Pval.adj))[,c(1,2,3,6)]
    colnames(Cor.table) <- c("gene1","gene2","cor","p.adj")
    Cor.table
    Cor.table.filt <- Cor.table [(abs(Cor.table[,3])>0.8),]
    write.table(Cor.table.filt, "Cor_comp1_0.8_gse7803.txt", sep="\t", row.names=F, quote=F)`

enter code here`

[Kevin Blighe]

What is the purpose of these lines:

 Norm.interest.corr1$p[lower.tri( Norm.interest.corr1$p,diag=TRUE)]=NA
 Pval.adj <- as.data.frame(as.table(Norm.interest.corr1$p))
 Norm.interest.corr1$r [lower.tri( Norm.interest.corr1$r,diag=TRUE)]=NA
 Correlation <- as.data.frame(as.table(Norm.interest.corr1$r))

By converting all of those values to NA, it makes sense that your data would be in only 1 direction.

Please see a quick tutorial, here: Network plot from expression data in R using igraph

[au.rinki.bio]

I am sharing a link for a better explanation, from where I get the code for construction co-expression network http://virtualplant.bio.puc.cl/share/step-by-step/step-by-step.txt

[Kevin Blighe]

Sorry, I have minimal time these days. When following a tutorial, though, you should check the output of every command to see what the command is doing. From what I can see, it makes little sense to run these commands (above) for the purpose of generating any standard signed (positive and negative) network.

[au.rinki.bio]

for the co-expression network construction, we use normal tissue sample and cancer sample at the same time. like we have 10 normal samples and 10 cancer samples. in the input matrix, we have a total of 20 samples in the column and corresponding genes in rows. after normalization of data used above-mentioned code for the co-expression network construction. please suggest a possible way if I am in the wrong direction.

Answer 1

[entropy]

10 is too small for correlations. you probably will not get healthy scores.