A unified drug–target interaction prediction framework based on knowledge graph and recommendation system

Abstract

Prediction of drug-target interactions (DTI) plays a vital role in drug development in various areas, such as virtual screening, drug repurposing and identification of potential drug side effects. Despite extensive efforts have been invested in perfecting DTI prediction, existing methods still suffer from the high sparsity of DTI datasets and the cold start problem. Here, we develop KGE_NFM, a unified framework for DTI prediction by combining knowledge graph (KG) and recommendation system. This framework firstly learns a low-dimensional representation for various entities in the KG, and then integrates the multimodal information via neural factorization machine (NFM). KGE_NFM is evaluated under three realistic scenarios, and achieves accurate and robust predictions on four benchmark datasets, especially in the scenario of the cold start for proteins. Our results indicate that KGE_NFM provides valuable insight to integrate KG and recommendation system-based techniques into a unified framework for novel DTI discovery.

MGraphDTA: deep multiscale graph neural network for explainable drug–target binding affinity prediction

“…These results suggest that it may be enough to use simple feature-based methods like RF in this scenario, which is consistent with a recent study. 64 Since the number of drugs in the Davis dataset is significantly less than that in KIBA and Metz datasets as shown in Table 1 , the generalization ability of a model trained on limited drugs can not be guaranteed for unseen drugs. Fig.…”

Target-specific compound selectivity for multi-target drug discovery and repurposing

“…However, the approach can be further developed by incorporating other information of either compounds or targets, for example, compound structural similarity ( Lo et al, 2019 ) to infer selectivity of novel compounds, even without any measured bioactivities. Alternatively, machine learning methods can be used to predict bioactivities for the compound-target pairs that have not yet been explored experimentally ( Bora et al, 2016 ; Merget et al, 2017 ; Öztürk et al, 2018 ; Thafar et al, 2019 ; Vamathevan et al, 2019 ; Bagherian et al, 2020 ; Nguyen et al, 2020 ; Schneider et al, 2020 ; Cichońska et al, 2021 ; Ye et al, 2021 ), after which the target-specific compound selectivity metric can be applied to the fully predicted compound target interaction matrix to identify selective lead compounds against any target of interest. In the general method development, we did not distinguish between the on- and off-targets, or penalized targets that may lead to adverse effects in clinical applications, but such factors could be later incorporated into the general selectivity scoring approach when applied to a particular disease or cellular context, similar to the KInhibition Selectivity Score ( Bello and Gujral, 2018 ), but this will require careful distinction between the therapeutic and toxicity-related targets.…”

Knowledge distillation for multi-depth-model-fusion recommendation algorithm

“…In order to alleviate the data sparsity and cold start problems in recommender systems, Ma et al [ 30 ] constructed a knowledge graph and made recommendations through various information such as user-item, neighbor-neighbor, etc. Ye et al [ 31 ] obtained low-dimensional representations of various entities by constructing a knowledge graph, and then input them into a neural decomposition machine for recommendation. Since traditional recurrent neural networks can only rely on linear transformations in each session to train recommendation models, Gwadabe et al [ 32 ] propose a graph neural network-based recommender system that simultaneously uses non-sequential interactions and sequential The interactive information is used for model training, which improves the model recommendation effect.…”