A statistical framework for revealing signaling pathways perturbed by DNA variants

Abstract

Much of the inter-individual variation in gene expression is triggered via perturbations of signaling networks by DNA variants. We present a novel probabilistic approach for identifying the particular pathways by which DNA variants perturb the signaling network. Our procedure, called PINE, relies on a systematic integration of established biological knowledge of signaling networks with data on transcriptional responses to various experimental conditions. Unlike previous approaches, PINE provides statistical aspects that are critical for prioritizing hypotheses for followup experiments. Using simulated data, we show that higher accuracy is attained with PINE than with existing methods. We used PINE to analyze transcriptional responses of immune dendritic cells to several pathogenic stimulations. PINE identified statistically significant genetic perturbations in the pathogen-sensing signaling network, suggesting previously uncharacterized regulatory mechanisms for functional DNA variants.

Reconstructing the Molecular Function of Genetic Variation in Regulatory Networks

“…To assess GEVIN's ability to identify the particular network branch in which molecular components are perturbed by a given reQTL, we compared GEVIN to two previously published methods: (i) The INCIRCUIT algorithm (Gat-Viks et al 2013), a qualitative method whose output is a list of branches hypothesized to be perturbed by an input reQTL based on the network structure, the output of which is not accompanied by any statistical assessment; and (ii) the PINE algorithm (Wilentzik and Gat-Viks 2015), a statistical framework that relies on probabilistic graphical modeling (Koller and Friedman 2009) of the regulatory network. PINE has been shown to outperform the INCIRCUIT algorithm (Wilentzik and Gat-Viks 2015) but is limited to homozygous organisms. All reported results were generated using the optimal parameters of each algorithm as described in Wilentzik and Gat-Viks (2015) (PINE: 500 permutations and a confidence level of 0.95; and INCIRCUIT: an association cutoff of 0.1 and enrichment cutoff of 0.9).…”

“…To apply GEVIN to these data we modeled the murine TLR/RLR signaling network (depicted in Figure 4A and described in Methods). We applied GEVIN to five groups of genes (groups #1 to #5) that were previously characterized in immune DCs (Gat-Viks et al 2013;Wilentzik and Gat-Viks 2015) and were scrutinized for the embedding of their associated reQTLs in the TLR/RLR signaling network. According to GEVIN's predictions ( Figure 4B and Figure S7 in File S1), two reQTLs (located in chromosome 1: 1282185 Mb and chromosome 9: 1222123 Mb) perturbed two distinct branches in the TLR/RLR signaling network (poly I:C-TRAF3 and LPS-TLR4, respectively), leading to genetic variation in the response of genes in groups #1 and #4, respectively (permutation-based FDR , 10%).…”

“…An alternative strategy considers the contextual specificity of both the reQTLs and the regulatory rewiring of signaling pathways. In two recently devised methods- INCIRCUIT and PINE (Gat-Viks et al 2013;Wilentzik and Gat-Viks 2015)we considered a mechanism in which a causal reQTL affects the transduction of signals along a certain molecular pathway (referred to as a "perturbed network branch"), thereby leading to a perturbed signal propagating along the network, from the extracellular stimulations toward the downstream transcribed genes (Supplemental Material, Figure S1 in File S1). In those methods, the reQTL mapping step exploits standard procedures for the identification of reQTLs, and the network embedding step searches for perturbed network branches by exploiting existing knowledge about the propagation of different stimulation cues through the regulatory network.…”

“…In those methods, the reQTL mapping step exploits standard procedures for the identification of reQTLs, and the network embedding step searches for perturbed network branches by exploiting existing knowledge about the propagation of different stimulation cues through the regulatory network. In particular, PINE was shown to improve on the performance of the INCIRCUIT algorithm, achieving substantially better predictive accuracy (Wilentzik and Gat-Viks 2015). However, analysis using PINE is still limited, most notably owing to its inability to handle heterozygosity and its high false-positive rate.…”

Network-based approaches for analysis of complex biological systems

“…Some approaches are applicable when there is a known directionality from the perturbation to specific affected targets. PINE (Perturbations in NEtworks) [84] is a probabilistic method to predict which branches of an input parameterized signaling network are affected by genetic variants under specific conditions. If a less detailed input network is sufficient, network flow methods (e.g., minimal set cover) can also be used.…”