Assessment of community efforts to advance network-based prediction of protein–protein interactions

Abstract

Comprehensive understanding of the human protein-protein interaction (PPI) network, aka the human interactome, can provide important insights into the molecular mechanisms of complex biological processes and diseases. Despite the remarkable experimental efforts undertaken to date to determine the structure of the human interactome, many PPIs remain unmapped. Computational approaches, especially network-based methods, can facilitate the identification of previously uncharacterized PPIs. Many such methods have been proposed. Yet, a systematic evaluation of existing network-based methods in predicting PPIs is still lacking. Here, we report community efforts initiated by the International Network Medicine Consortium to benchmark the ability of 26 representative network-based methods to predict PPIs across six different interactomes of four different organisms: A. thaliana, C. elegans, S. cerevisiae, and H. sapiens. Through extensive computational and experimental validations, we found that advanced similarity-based methods, which leverage the underlying network characteristics of PPIs, show superior performance over other general link prediction methods in the interactomes we considered.

HGNNPIP: A Hybrid Graph Neural Network framework for Protein-protein Interaction Prediction

“…A deep understanding of PPIs can provide global insights into cellular organization, genome function, and genotype-phenotype relationships in various species 69 . Increasingly, computational algorithms are being developed to discover previously unrecognized PPIs and, thereby, improve the comprehensiveness of the interactome map 1 . In this study, we proposed a Hybrid Graph Neural Network model (HGNNPIP) for Protein-protein interaction prediction.…”

“…The optimization of this model is to maximize the probability of the context word 𝑆[𝑖 ± 1] given a center word 𝑆[𝑖]. We then obtained the embedding matrix 𝐸 ∈ 𝑅 )*×, , in which the 𝑗-th row represents the feature vector of 𝑗-th amino-acid residue (𝑗 ∈ [1,20]). In another word, each amino-acid residue is embedded into a 𝑟 −dimensional space.…”

GraphGuard: Detecting and Counteracting Training Data Misuse in Graph Neural Networks

“…These edges are considered the property of the company, and they may express concern over the privatization of their graph structure data. Another example could be a GNN used for PPI in drug discovery, where nodes represent proteins and node attributes represent well-known properties [36,59]. The edges represent interactions among proteins, which should be obtained by extensive experimentation [40,44].…”

“…The data misuse issue also gains prominence in the context of GNNs since they are utilized in sensitive domain applications, such as using GNNs to predict properties in Protein-Protein Interaction (PPI) graphs in pharmaceutical research [33,34,43,47]. In such scenarios, nodes signify proteins and edges indicate their interactions [36,59]. The construction of these graphs demands extensive experimentation and significant financial investments, making them valuable intellectual property [40,44].…”

State of the Interactomes: an evaluation of molecular networks for generating biological insights

“…This approach leverages interaction prediction algorithms, which provide an efficient method for addressing network incompleteness and have previously been used to identify and prioritize novel interactions 16,17 . Previously, benchmarking studies have assessed the performance of different interaction prediction algorithms across a small number of human interactomes 18,19 . Based on these studies, we now implement interaction prediction across a wide range of interactomes to assess the influence of the underlying network architecture on prediction accuracy.…”