I have some data to make cluster analysis. Suppose these data are five columns; ID of gene, t-value, P.value, Adj.P.value and FC. I would like to cluster it and find a group of data showing similar P.value. Suppose I want a hierarchical clustering in dendogram. If I select ID, then it is showing a cluster with ID. My question is that " Is this cluster is formed on the basis of other parameters such as P.value or only based on ID". In addition I would like to get the right commands needed to get K-means clustering, PCA and hierarchical clustering in software R.
Hi Felix Seifert, Thanks, Your information is really very helpful. I would like to tell you little more details. I have done microarray of sample with two seperate conditions. I have extracted data and then find the fold change in comparing with WT. After that, the present data format has the following heading: ID of gene, t value, P.value, Adj.P.Value and FC. My intention is that if I make cluster analysis ( I am not expart in Biostatics), I can get cluster of the genes with similar change of gene expression level. To do this, I prefer P.value. Suppose, I want to know which sets of genes are showing similar fold (1.5-2) change within the P.value range of 0.001-0.005.
By this way, I would like to find the important gene/gene sets that are affecting by the condition I set.
I hope you get it. If you have any other suggestion, please inform me.
Clustering on p-value or adjusted p-value will simple tell you which genes displayed similar levels of statistical significance. It tells you nothing about which genes had similar changes in expression as two genes may have identical statistical significance and very different differential expression.
Your p-values are a function of the t-statistic (i.e. estimated log2 fold change divided by the standard error of the estimate) so they reflect differences in the variance about the mean expression amongst the replicates used to compute the log2 ratio for fold change. They do not really tell you anything useful about the actual log2 expression itself.
If you want to cluster genes by similar expression, then you need to use a measure of expression - either (replicate) mean normalized expression or, since it is simpler and you already have the values in hand, the log2 fold change estimates of differential expression for those genes.
Program for k-means clustering
library(simpleaffy)
library(hgu95av2.db)
library(annaffy)
setwd("/home/ibab/Systemsbio/Case3")
raw.data = read.affy("groupfile.txt")
x.rma = call.exprs(raw.data,"rma")
results = pairwise.comparison(x.rma, "Group", c("Normal", "hcc"), raw.data)
pc.f = pairwise.filter(results,tt=0.05,fc=1)
probeid = rownames(means(pc.f))
symbols = aafSymbol(probeid, "hgu95av2.db")
symbols = getText(symbols)
names(symbols) = rownames(means(pc.f))
pathway = aafPathway(probeid, "hgu95av2.db")
pathway = getText(pathway)
names(pathway) = rownames(means(pc.f))
fit
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Statistical Software
- R