Interactions between proteins are crucial for the correct functioning of the living cell. That is why it is important to study protein interaction networks, detect hubs, leaves and network modules and try to understand their biological roles.
It is known that current protein interaction datasets suffer from study bias (see for example http://gtr.rcuk.ac.uk/project/3CB73012-FFE1-49E3-921B-5DED875658A4, http://journal.frontiersin.org/article/10.3389/fgene.2015.00260/abstract or http://www.cell.com/abstract/S0092-8674(14)01422-6), i.e. the greater the number of reported interaction partners of a protein, the more it has been studied, mainly due to its role in cancer or other relevant topic in Biology (see Fig. 1a). This doesn't mean that the interactions measured for this protein are not real, it is only an indication that there are some less studied proteins, which could have more interactions that the ones observed in available datasets.
Given that the node degree distribution (i.e. the number of interaction partners of a protein) and the distribution of the number of publications/studies associated with a protein have heavy right tails (see Fig. 1b,c), how would you approach this bias correction problem?
In other words, how can we find a cut-off in the number of studies or protein degree to have a more reliable protein network?