Hi,
Do you know about any other gene gene interaction database other than (bisogenet's) sysbiomics? Thank you in advance.
You mention Sysbiomics. That is actually a meta database for protein-protein and protein-DNA interactions. (I think that also answers Neil's questions what kind of interactions you are looking for.) According to it's usermanual it contains: "Functional relations (protein-protein and protein/DNA interactions) ... from BIND[4], HPRD[5], Mint[6], DIP[7], BioGRID[8] and Intact [9])." (refs are in the [?]manual[?]. That is a whole lot, maybe even too much since some of these databases overlap. If you use those in Bisogenet (which is a Cytoscape plugin) your most likely problem is that you will find too many interaction and will end up with a classical hairball problem.
If that is indeed your problem you might want to work with the best curated subset. Now it might be hard to decide what is the best. But I would start with IntAct since it indeed is highly curated (it is fact based on BIND) and still actively maintained. In Cytoscape you could of course also label the interaction based on their original source and see what happens if you use specific subsets.
I have thrown one of your proteins (CAPN10) into IntAct. It indeed gives zero hits in IntAct itself but interestingly IntAct does give interactions from other databases as well. These are really provided through the Proteome Standard Intiative PSIQuic webservice). For CAPN10 it found 129 interactions 127 of these from STRING (mostly derived from text mining, STRING contains about 100 times as many interactions as most other databases listed there). You might try to evaluate some of your more important proteins manually in this way and use the information you get to decide what other databases would be useful. This also seems to show that your problem probably is not an identifier problem, so my first comment below might be less relevant.
It would be interesting to know the real biological problem you try to answer.
(Sorry if this is more a comment than an answer, it did not fit in the comments easily)
My problem is the underrepresentation of the genes in the network when I use the sysbiomics. The most importnat genes do not show any interaction, though that gene could be found when checking the sysbiomics database. That is why I wanted an accurate database which shows gene-gene interactions. Some of such are FTO, SLC30A8,CDKAL1,CAPN10. This study is on Type 2 Diabetes Mellitus.
So you are convinced that some of your proteins should be in the database, but they are not. Since you say they do not show any interactions your problem might in fact be a identifier mapping problem. I could imagine that your proteins actually are in the database, but with an ID different than the one you are using. It would be worth checking the Sysbiomics and Bisogenet manuals to see what identifiers they expect.
maybe you should branch further away from what are the currently known interaction datasets. (For me the hairball 'problem' isn't a problem). If your genes are underrepresented it could be that they don't bind many things, that they haven't been focussed on (unlike the many false hubs) or that they don't work particularly well in the high-thru assays that end up in Mint/BioGrid etc (or all of the above for the SLiCks). Try http://hanlab.genetics.ac.cn/sys/intnetdb for computationally predicted interaction partners
While computationally predicted interactions (or derived from co-citation based textmining for that matter) can be very interesting as a starting point, they are also just what they are, predictions or things that happen to appear in the same context. They need to be experimentally verified and that verification data needs to be curated. RussH is right thought that what really is verified and curated is biased to what happens to be interesting. So yes using massive amounts of predicted interactions can be a nice start, but better work with a good wetlab to do the verification.
A gene-gene interaction is when genetic variations or markers in two different and distinct loci contribute together to a phenotype. A variant in one gene without the second does not give the phenotype. This is also called epistasis.
For protein-protein interactions, which seems to be the topic here, I use data from BIND, BioGRID and HPRD - links provided should be for glucose transporter SLC2A2. Yes, FTO has no known or verified protein-protein interactions. The hype was all about obesity and genetics for FTO and mouse Fto with little done on its other functions. Now, we know more:
FTO catalyzes the demethylation reaction of 3-methylthymine (ssDNA) + M Fe(II) + 2-oxoglutarate to succinate + formaldehyde + CO2 (Gerken Schofield 2007 Science)
A dominant mutation (resulting in p.Ile367Phe) in the mouse Fto gene that reduces its DNA demethylation activity also results in reduced fat mass (Church 2009 PLoS Genet 5:e1000599)
So, perhaps by mining the literature, as I have done here, you can add some of your own interactions to those from the three sources mentioned above. Such enhancements will make your stand out compared to downloading what everyone else has. OPne interesting interaction could be from Fto or FTO to the phenotypes associated with this gene.
Use of this site constitutes acceptance of our User Agreement and Privacy Policy.
version 2.3.6
Gene-gene interaction can mean many things - can you clarify what you mean by the term? For example, some people use it (incorrectly) to refer to physical interaction between protein products of the genes. Genes with similar microarray expression profiles are also candidates for "interaction".
Indeed it is the protein protein interaction in actual terms. The genes involved in creating those proteins are needed to be consider here.