This really depends on the biological question you want to answer.
A typical basic scenario would be: You have a functionally annotated genome (e.g. with Blast2GO) and identified a list of up-regulated genes under a certain condition.
Your question would be: Is there a certain molecular function overrepresented among these genes?
Technically speaking: You take the GO annotations of the whole genome (reference set) and a list of upregulated genes (test set) and perform a Fisher's Exact Test to find statistically overrepresented functions. The test compares all functional associated to the genes in your list to the ones of the rest/whole genome. Since this test is done for each function (e.g. a GO term) separately you have to adjust for multiple testing via e.g. a false discovery rate (FDR). The result is a list of functions with an adjusted p-value below (or close to) 0.05. Blast2GO allows to further summarize, visualize, condense and export these functions.
This really depends on the biological question you want to answer.
A typical basic scenario would be: You have a functionally annotated genome (e.g. with Blast2GO) and identified a list of up-regulated genes under a certain condition.
Your question would be: Is there a certain molecular function overrepresented among these genes?
Technically speaking: You take the GO annotations of the whole genome (reference set) and a list of upregulated genes (test set) and perform a Fisher's Exact Test to find statistically overrepresented functions. The test compares all functional associated to the genes in your list to the ones of the rest/whole genome. Since this test is done for each function (e.g. a GO term) separately you have to adjust for multiple testing via e.g. a false discovery rate (FDR). The result is a list of functions with an adjusted p-value below (or close to) 0.05. Blast2GO allows to further summarize, visualize, condense and export these functions.
Hi, you may conduct significance test separately for each GO set (BP, MF and CC). The choice is yours but generally Fisher's single tail t-test (exact t-test) with FDR correction is considered good enough for over representation analysis. If you consider all the three GOs together, then you are including the genesets which may be irrelevant to your analysis. In many cases, for plant gene over representation analysis CC genesets may not mean much...