PICKLE 2.0: A human protein-protein interaction meta-database employing data integration via genetic information ontology

Abstract

It has been acknowledged that source databases recording experimentally supported human protein-protein interactions (PPIs) exhibit limited overlap. Thus, the reconstruction of a comprehensive PPI network requires appropriate integration of multiple heterogeneous primary datasets, presenting the PPIs at various genetic reference levels. Existing PPI meta-databases perform integration via normalization; namely, PPIs are merged after converted to a certain target level. Hence, the node set of the integrated network depends each time on the number and type of the combined datasets. Moreover, the irreversible a priori normalization process hinders the identification of normalization artifacts in the integrated network, which originate from the nonlinearity characterizing the genetic information flow. PICKLE (Protein InteraCtion KnowLedgebasE) 2.0 implements a new architecture for this recently introduced human PPI meta-database. Its main novel feature over the existing meta-databases is its approach to primary PPI dataset integration via genetic information ontology. Building upon the PICKLE principles of using the reviewed human complete proteome (RHCP) of UniProtKB/Swiss-Prot as the reference protein interactor set, and filtering out protein interactions with low probability of being direct based on the available evidence, PICKLE 2.0 first assembles the RHCP genetic information ontology network by connecting the corresponding genes, nucleotide sequences (mRNAs) and proteins (UniProt entries) and then integrates PPI datasets by superimposing them on the ontology network without any a priori transformations. Importantly, this process allows the resulting heterogeneous integrated network to be reversibly normalized to any level of genetic reference without loss of the original information, the latter being used for identification of normalization biases, and enables the appraisal of potential false positive interactions through PPI source database cross-checking. The PICKLE web-based interface (www.pickle.gr) allows for the simultaneous query of multiple entities and provides integrated human PPI networks at either the protein (UniProt) or the gene level, at three PPI filtering modes.

Model‐based detection and analysis of introgressed Neanderthal ancestry in modern humans

“…We also tested if proportion of Neanderthal ancestry was positively correlated with local population-scaled recombination rate (as inferred by Myers et al (2005)), which would be suggestive of selec- If DMIs were the cause of selection against introgression, then we would expect that genes that code for proteins with more binding partners would be less likely to be introgressed; each proteinprotein interaction (PPI) can be thought of as a possible DMI. To test this hypothesis, we used the PICKLE2.0 PPI network (Gioutlakis, Klapa, & Moschonas, 2017;Klapa, Tsafou, Theodoridis, Tsakalidis, & Moschonas, 2013) and associated a number of binding partners to each gene by counting the number of PPIs in which the protein coded by that gene participates. We found an insignificant correlation between number of binding partners and mean frequency of introgression in both populations (Spearman's q = À.016, p = .314 in CEU, q = À0.003, p = .858 in CHB + CHS, two-sided bootstrap-like test), which provide weak indirect evidence against the DMI hypothesis.…”

Model-based detection and analysis of introgressed Neanderthal ancestry in modern humans

“…If DMIs were the cause of selection against introgression, then we would expect that genes that code for proteins with more binding partners would be less likely to be introgressed; each protein-protein interaction (PPI) can be thought of as a possible DMI. To test this hypothesis, we used the PICKLE2.0 PPI network (Klapa et al, 2013;Gioutlakis et al, 2017) and associated a number of binding partners to each gene by counting the number of PPIs in which the protein coded by that gene participates. We found an insignificant correlation between number of binding partners and mean frequency of introgression in both populations (Spearman's ρ = −0.002, p = 0.915 in CEU, ρ = 0.007, p = 0.648 in CHB+CHS, two-sided bootstrap-like test), which provides indirect evidence against the DMI hypothesis.…”

Cell type-specific proteogenomic signal diffusion for integrating multi-omics data predicts novel schizophrenia risk genes

“…The overall structure of MAPSD is provided in Figure 1. MAPSD starts with a large-scale protein-protein interaction (PPI) network which was assembled from multiple sources [30][31][32][33] (see Experimental Procedures). Using the PPI network, an affinity matrix is created.…”

“…Interaction networks used in this study were collected from three sources including: PICKLE 2.3 32,33 , The Human Reference Interactome 31 , and human Interactome Database 30 . Upon removing the duplicate interaction, the final network being used by MAPSD contained 232,801 interactions.…”