Be warned. DAVID is no longer supported and is now out-of-date. It was last updated in 2010 so any annotations to source databases since then are not in DAVID:

http://david.abcc.ncifcrf.gov/forum/viewtopic.php?f=10&t=807

In my experience with DAVID (http://david.abcc.ncifcrf.gov/), it is best to focus not on P-values, but on the annotation of the genes, and their possible clinical or biochemical relevance. I use DAVID strictly for variant prioritization based on clinical/biochemical relevance using DAVID's functional annotation and pathway analysis. I do not separate up- or down-regulated genes when I run genes from an expression analysis, and don't worry about DAVID P-values, because I have previously determined the genes with significant P-values. Also, I study patients with immune disorders, so I am only interested in genes involved in immune/cancer/apoptosis/DNA repair and certain other pathways that I am very familiar with. Genes in these pathways are often not the genes that cluster in with the lowest P-vales in DAVID. If you know something about the patient or cellular phenotype, use a "clinical eye" to scan DAVID results both before and after functional annotation clustering, and good candidates will sometimes just pop out at you. Remember to click on the blue bars for detailed information throughout the DAVID analysis (I don't think this is obvious when you first start using it). Good luck!

Thanks for your answer, how many genes (500,1000,2000,...!!.do you enter to  David per analysis?

I think the limit for DAVID is 3000, but I usually have fewer than that as I have already minimized the list based on differential gene expression. I should note though, I am currently analyzing targeted RNASeq data, not whole transcriptome. 

If you have human/mouse/rat, try GeneGo or Ingenuty. Commercial, but because of that have better databases (human-handpicked=curated) than DAVID, which has its database very much from automatized KEGG bits...

As for the number of genes, for pathway/functional the enrichment I started (in geneGo) with as little as 40-50 (differentially expressed) to get some sort of informative results. Using >2000 as differentially expressed may give on output mostly artefactual results - the pathways that are most popular/most fully annotated. Optimum was typically 100-400.

Be warned. DAVID is no longer supported and is now out-of-date. It was last updated in 2010 so any annotations to source databases since then are not in DAVID:

http://david.abcc.ncifcrf.gov/forum/viewtopic.php?f=10&t=807

I thank you for your help. Actually I started doing the analysis with GSEA program : http://www.broadinstitute.org/gsea/index.jsp. We had some confusing results with, so we looked for another method and we applied genespring GS Agilent and the obtaining genes lists from the interpretations, that we did, were treated with DAVID program where I had some questions. Now,I started to do the analysis with  Ingenuity IPA. thanks

I agree, Ingenuity is great if you have access to it. Christian's comment about DAVID being out of date and no longer supported is well taken, but as DAVID links out to other databases that are well supported, such as Genecards, it is still an awesome tool for quickly prioritizing gene lists based on function.

DAVID program helped me in the beginning  specially that I did not do this type of analysis before Our obtained results by DAVID will be compared with what we get from Ingenuity and a general interpretation (discussion) of the results from DAVID, GSEA and Ingenuity in the end.

Hello Iman,

What kind of gene-expression data are you looking at, Is it from RNA-Seq or Microarrays? You might want to look at the background you've set in david, if its microarray i suggest you to select the specific platform you are using in the background tab, or upload a list of background probes if they're not available. The default background is just all genes in the genome. However, this option should be carefully considered because it is often source of mistakes. If your background in enriched for phosphoprotein coding genes, then this might explain why you see them being significant in both conditions.

Also I suggest you to be careful with the online version of GSEA as in the investigate gene-sets option you don't have the opportunity to select the proper background. If you have the opportunity, I would suggest you to download the R-GSEA package along with all the gene-sets collections and then compute the hypergeometric statistical tests from within R and set the correct background.

However, if you look for an online-based tool you may also want to have a look at WebGestalt (http://bioinfo.vanderbilt.edu/webgestalt/), which used to be a great tool and a great first choice alternative to David. Last website update was mid-2014.

Thank you Marco. My gene-expression data are from microarray. It is true that I did not looked at the background which is Mus musculus in my experiment. When I do analysis with DAVID:  I do upload of my data and then I choose  the identifier (i.e. Agilent oligo) and gene list  before submit list. I am actually concerning on doing the analysis via IPA ingenuity. What was useful for me and avoided  to see a lot of molecular functions (ion binding, cation binding,.....etc) is that I reduced the number of genes before being treated with DAVID.