I have recently carried out a microarray experiment with human cell lines (Affymetrix, S Clariom chips). I now have the top gene lists and want to carry pathway and biological process analyses. I have asked experts on the field about which genes I should include for these analyses but I get contradictory answers. My options are as follows:
1) Include the top 45 genes which differ by Fold change > 1.5 and p0.05).
2) Include 4,000 genes which have a p
Hi Dimitra,
This is my opinion:
If you have done your statistical analysis right (you have corrected for multiple hypothesis tests), then it means that most genes in your list are indeed enriched. The expression of some genes simply does not change so dramatically.
Therefore, I would include all significant genes. Of course, there will be some false positive genes in your list but the process of pathway enrichment is also statistically affirmed. Unless your random false positive hits don't randomly occur in the same pathway your results should be reliable. You can always use a more stringent FDR for the enrichment.
Ps. remember that you should do your pathway analysis against a "background" list of genes. The background should be the list of all the genes that you measured - the genes that had the chance of coming out as significantly enriched. David functional enrichment (online tool) allows you to upload a list of significantly enriched genes as well as a background list.
Dear Dimitra Makri
I agree with Jan Zaucha that you should only include your most significant genes. However, I also have a suggestion.
Make a Gene Ontology analysis using more than one tool. After 11 years of performing GO analyses, I saw that different tools provide distinctly results that can be so heterogeneous that becomes difficult to isolate the best answer. However, if you perform the analysis in multiple software, you can see the whole panorama. The best GO tool I recommend is topGO for R. But if you want a strictly online tool, DAVID might be the best option. KEGG can work too but depend on your organism and final goal. Also, be careful of which statistical tool you choose. For example, the famous Benjamini-Hochberg FDR is less rigorous than Bonferroni Family-Wise for GO. If you need further guidance, please write again.
Cheers.
If you want to get your hands dirty, you could perform the statistical analysis using the R programming language(or better ask a bioinformatician).
There are various packages for the analysis of microarray data such as affy (https://bioconductor.org/packages/release/bioc/html/affy.html) and limma(https://bioconductor.org/packages/release/bioc/html/limma.html).
On the vignettes you can find detailed ways to perform the analysis and there is always the https://support.bioconductor.org/t/Latest/ help site to ask questions.
As Bruno César Feltes and Jan Zaucha suggested already there are ways to perform functional enrichment analysis using R or web tools.
I would like to add Enrichr as a web tool (http://amp.pharm.mssm.edu/Enrichr/) because it performs the analysis in a lot of different databases that can provide you with more insight about your results.
Dear Jan Zaucha Bruno César Feltes and Konstantinos Geles ,
many thanks for your thorough and consice replies. They have helped me choose the best way to follow up the microarray results.
Indeed, as Jan Zaucha highlighted, the original analysis via Partek has a Benjamini-Hochberg correction for detecting false positives. Therefore, I will follow your advice and add the top gene list with p
- Badges
- Science method
- Similar topics
- Mathematics
- Statistics
- Statistical Modeling
- Analyses of Variance
- ANOVA