How to interpret terms with “positive” and “negative” specifiers

Question

[chriselhadi]

I am currently interpreting some GO terms I’ve clustered using binary clustering. The problem is that in every cluster, I get tens of terms having specifiers as “positive” and “negative regulation of …”. So I’m really confused and can’t decide how to interpret the results. What I did is the following: I ordered the terms according to their BH adjusted p values, and then decided to consider the specifier having the most significant p-value as ultimately describing the regulation of the cluster’s function… Is that the right way to do it? And what about the other specifier, should I just ignore its presence in favor of the more significant (e.g if the term “negative regulation of …” has the lowest pvalue, regardless if there is other “positive regulation of …” terms with higher and less significant pvalues, should I just assume the regulation is negative and ignore the positive terms?)? I also thought about counting the positive and negative terms and then take the specifier having the largest count as describing the cluster. Am i thinking right?

Answer 1

[Kevin Blighe]

Genes can be multifunctional and be part of multiple pathways; so, false-positive enrichments in in silico analyses will always occur. Simply taking the most statistically-significantly enriched term is not necessarily going to ameliorate the problem.

If you look in published works, you'll see that even in major publications there sometimes appear pathways / enrichment terms that also seem out of place. So, this problem of false-positive enrichment is known / understood, and tolerated.

What you could be doing is setting a p-value threshold, possibly coupled with some other thresholds like enrichment score and min features / genes per term, and then examining all terms that pass these thresholds. It is then up to you to interpret which is / are most relevant, and this can only really be done by having an understanding of your study design and disease area.

For example, if I am studying asthma and enrich for Hepatocyte, Airway Epithelium, and Dexamethasone Signalling, then obviously I will assume that Hepatocyte is a false-positive, while the other 2 are not.

Kevin