A hierarchical method for finding interactions: Jointly using linear correlation and rank correlation analysis

摘要

In the earlier studies, I pointed out that a network changed in a local domain can be approximated as a linear network, i.e., all between-node (or -taxon, -component, etc) changes in the local domain are treated as linear ones and Pearson linear correlation measure can be used. For a little wider domain, the quasi-linear measure, Spearman rank correlation can be used also. In present study, I jointly use Pearson linear correlation measure and Spearman rank correlation measure and their partial correlations to find interactions. First, I define some hierarchical principles for finding interactions. Reliability levels are then defined using set operations. The full algorithm and Matlab codes for finding interactions are given.