I don't think there's an objective list of cancer-associated GO terms. I would take a list of known cancer genes (like this) and look for enriched GO terms among these genes to get cancer-associated GO terms.

I think you need to be aware of what the mission of GO is. The strategy (at least as I was aware of it during earlier developmental stages of the consortium) was to capture "normal" biology. There was a major conflama once when they had a term--I think it was "tumor suppressor"--but that was moved to "obsolete" for exactly that reason.

Here's what I mean: http://amigo.geneontology.org/cgi-bin/amigo/term_details?term=GO:0008181&session_id=1456amigo1307459582

This term was made obsolete because it does not represent a true biological function, rather a pathology that occurs when a particular gene product is inactivated in some way.

This may have changed, I'm not as in touch with the daily GO machinations now. But it's a philosophical distinction about the ontology that people might be interested to know. You might want to use a different ontology I guess is what I'm saying.

You might want to check this paper.

You can use the list of genes under C4: Computational gene sets compiled in MSigDB as an initial dataset for your analysis. Data is available in 2 sections as CGN: cancer gene neighborhoods ( 881 gene sets) and CM: cancer modules (427 gene sets).

You can also check the gene-family classification in GSEA for a set of gene categorized in to oncogenes, tumor suppressors etc.

You may use the GO terms from these gene sets for a library of cancer related biological processes.