Network link prediction by global silencing of indirect correlations

Abstract

Predicting physical and functional links between cellular components is a fundamental challenge of biology and network science. Yet, correlations, a ubiquitous input for biological link prediction, are affected by both direct and indirect effects, confounding our ability to identify true pairwise interactions. Here we exploit the fundamental properties of dynamical correlations in networks to develop a method to silence indirect effects. The method receives as input the observed correlations between node pairs and uses a matrix transformation to turn the correlation matrix into a highly discriminative silenced matrix, which enhances only the terms associated with direct causal links. Achieving perfect accuracy in model systems, we test the method against empirical data collected for the Escherichia coli regulatory interaction network, showing that it improves on the best preforming link prediction methods. Overall the silencing methodology helps translate the abundant correlation data into valuable local information, with applications ranging from link prediction to inferring the dynamical mechanisms governing biological networks.

Inferring functional connectivity in fMRI using minimum partial correlation

“…Its application is illustrated using a resting-state fMRI dataset from the human connectome project [24] . Results with the elastic PC-algorithm were compared with full correlation, fully partial correlation, regularized inverse covariance (ICOV) [16,17] , network deconvolution (ND) [37] and global silencing (GS) [38] . A brief description of these methods is as follows: Step 1 with the significance threshold α.…”

Newton’s Gravitational Law for Link Prediction in Social Networks

“…However, application of link prediction is much wider than in the above example. Link prediction has also been used to predict potential collaborators in business [32], in shopping recommendation [28], in patent partner recommendation [46], in predicting cell phone contact [38], in gene expression networks [7], in protein-protein interactions [2] and, in security related domain to detect suspicious communication [31].…”

Network Dynamics as an Inverse Problem

“…Whether one studies gene regulatory networks, metabolic networks or neural networks, sometimes we may have no means of directly measuring the network connectivity [20,21,[33][34][35][36][37][38][39]. Instead, one is forced to resort to indirect approaches to estimate the network connections from available data.…”

“…For instance, spike trains of neural networks may be used ti reveal network connections between spiking neurons [34,35,40]. Moreover, gene expression levels may also reveal the regulatory features of gene regulatory networks [4,7,9,20,21,33,37]. However, when inferring the connectivity of networks, one may stumble upon two basic types of connectivities, functional and structural connectivities.…”

“…Some examples of these measures are mutual information [41], transfer entropy [42], Granger causality [43,44] and extensions thereof. Although approaches of this nature are widely abundant in the literature [37,[45][46][47][48][49], they generally show a series of conceptual challenges when used to infer network connectivities. For instance, the functional connectivity may depend on the current network state, therefore, a given physical network may span different functional connectivities for different collective states [11].…”