RNA-SEQ: Gene ontology terms is vague

Question

[Muha0216]

hi i am looking at a particular treatment on global gene expression pattern of breast cancer cells. I have 3 questions with regards to the gene ontology terms.

1) One of the reasons why we also want to look at rna-seq level of data is apart from wanting to study mechanosensitive genes (genes which expression is altered upon mechanical force), we also want to study any interesting changes to genes implicated in cancer hallmarks (apoptosis, metastasis, proliferation, angiogenesis etc). But the GO terms are vague such as 'intracellular component'. Is there an online tool where we can input in all DEGs and classify each in terms of cancer hallmarks (genes involved in apoptosis, genes involved in metastasis etc)?

2) Q value low: some of the Q value for GO terms are HIGH, suggesting it is not significantly enriched. In terms of data interpretation, i cant say that this GO term is enriched but i cant deny the fact that there are DEGs involved in the particular GO term. Therefore can i still make some mention that for example, ''we found genes to be upregulated belonging to cell adhesion processes and they are gene A, gene B, gene C suggesting that the mechanical force might affect biological process A etc''??

3) Q value cutoff. what is a good Q value cutoff? I have often seen papers using <0.05 but a colleague of mine working on bioinformatics suggest that people in their lab used <0.1 as cutoff.

Answer 1

[jared.andrews07]

1.) How are you doing your enrichment? This will help us better interpret Q2 and Q3 as well. I like tools like enrichR, DAVID, etc, that make pathway enrichment very straightforward and usually have pretty good default settings in terms of their thresholds. And lots of GO terms are vague - that doesn't mean you have to comment on them.

You can use also try GSEA to compare your DEGs to known sets of genes involved in different pathways. There are tons of sets related to cancer hallmarks (and likely some for mechanosensitive genes as well). It's very commonly used and likely worth your time.

2.) This is a stretch in my eyes. Though often a flawed mindset, people like a p-value (or q-value or whatever statistic) that backs up whatever claim you're trying to make. If you have experimental evidence to back it up, that's a different story.

3.) Really depends on the application. I'd personally never believe a pathway enrichment with a q-value of 0.1 though. Be sure not to relax your thresholds for the sole purpose of trying to include the things you want to show. I'd stick with 0.05 at a bare minimum.

[Muha0216]

Hi jared. Thanks for your valuable input. How is the GSEA analysis different from a KEGG pathway analysis? We have one pathway analysis using the KEGG analysis.