Search for clusters among co-occurring objects and detection of protein complexes

Abstract

The method of statistically based search for structures in data of different natures, previously proposed by the authors, is applied to the problem of searching for clusters of objects in sets of co-occurring objects. In bioinformatics, a similar problem arises when identifying protein complexes. A method for identifying significant interactions between proteins is proposed, based on the use of a statistical model in which the corresponding p-values are calculated. The algorithm for searching for clusters in one-dimensional data allows us to identify a cluster of small p-values located in the vicinity of zero. Next, we consider protein complexes that are formed only taking into account these significant interactions. Among the resulting complexes, as a rule, a giant component appears which in turn can be clustered. The proposed method for identifying significant deviations from the null hypothesis, based on searching for clusters of p-values in the vicinity of zero, is general.