Gene Ontology: Disjoint Terms

Question

[Lorrit]

Hi all,

I recently started working with the Gene Ontology (GO), a database which allows to classify genes by their type or function. Each gene can be classified to several GO terms. My question is, whether there is some data available about disjoint GO terms, e.g. if gene X is assigned to GO term A, it cannot belong to GO term B.

I found some initial work here and a small data set here, but it is pretty small, quite old and seems to be concerned only with cellular components. Does anybody know further sources or larger databases containing information about which GO terms are disjoint?

Thanks for your help.

Lorrit

[Lorrit]

no ideas or suggestions?

[Jean-Karim Heriche]

Please don't use the "add answer" field for something that doesn't answer the question. It marks your question as answered so people would could otherwise answer may skip it (For my part, I don't often look at already answered questions)

Answer 1

[Jean-Karim Heriche]

The solution is to build a slim ontology that suits your needs and define an order for term assignment. For example if a protein has a kinase domain and an SH3 domain, you have to decide which one is more important/relevant to your biological question and assign that term to the protein, discarding the other. I've done this in the past when dealing with a specific biological process when I wanted to know which genes were known to be associated with the process of interest versus a list of some other processes.

[Lorrit]

Thank you for your answer. If I understand it correctly, you suggest to pick one (relevant) GO term for a protein that is assigned several terms. My question, however, aims towards finding unknown links between proteins and terms. For this task, I need information telling me "if protein X already is a term A, it cannot be a term B". Do you know any source containing this data?

[Jean-Karim Heriche]

Finding unknown links between proteins and terms can be seen as a data-driven prediction task to which you can apply a number of statistical learning approaches. In general, you don't need to know in advance "if X is A, it can't be B". Instead, you could get this information from a model you've learned from the data.
For an example of one possible approach to make predictions, see this paper but there may be better ways for your particular problem.