Entering edit mode
5.9 years ago
Swimming bird
▴
20
Hello! I am working with a protein-protein interaction network downloaded from iREFINDEX with only human-human interactions. Is it possible to obtain a confidence/reliability normalized score of each interaction using some software or web-based tool? iREFINDEX shows only some values regarding to PubMed identifiers.
My idea was to use MIscore but it is currenly discontinued.
IRefIndex integrates protein-protein interactions from multiple sources to give a more comprehensive network than can be obtained from using only one source. To associate confidence scores to interactions, you would have to define what is a good measure of confidence for you and use additional information to compute it. Some resources already do this but different people use different scores. Confidence in a reported interaction also depends on many contextual information not captured by the databases. In my opinion this kind of score is pretty useless unless it is tailored to a particular situation and in many cases not necessary. It often amounts to reinforce biases in the data (e.g. the most studied proteins/genes have the most interactions usually with the highest scores)
Could you recommend any tool easy to use? In my case I am interested in experimental interactions, in other words, those interactions inferred from literature or predicted are not useful for my analysis.
I typically write my own scripts for this kind of tasks. At least MIPS, IntAct and BioGRID collect information on the experimental methods used to reveal each interaction. You could combine the data from these databases in the same way as iRefIndex does while preserving the experimental method info or you could start from iRefIndex and trace back interactions to these three databases and extract the experimental methods associated with each.
In fact, iRefindex shows the experimental methods for each interaction. How can I obtain a score using this info?
Using the type of experiment alone, all you can do is count the number of methods by which an interaction is found or come up with an arbitrary scoring system based on your belief about relative reliability of the different methods. Neither of these is very satisfactory (e.g. interactions with oncogenic proteins are found with many methods so tend to have a high score because people desperately need to find a link between what they study and cancer to get funding). I would simply use the interaction graph as is (after all, if an interaction is published, it must be true). Otherwise depending on what you need the score to reflect, you could/should use additional information.