Abstract:Background: The potential of graph neural networks (GNNs) to revolutionize the analysis of non-Euclidean data has gained attention recently, making them attractive models for deep machine learning. However, insufficient compound or moleculargraphs and feature representations might significantly impair and jeopardize their full potential. Despite the devastating impacts of ongoing COVID-19 across the globe, for which there is no drug with proven efficacy that has been shown tobe effective. As various stages of drug discovery and repositioning require the accurate prediction of drugtarget interactions(DTI), here, we propose a relational graph convolution network using multi-features based on the developed drug chemicalcompound-coronavirus target graph representation and combination of features. During the implementation of the model, we further introduced the use of not only the feature module to understand the topological structure of drugs but also the structure of the proven drug target (i.e., 3CLpro) for SARS-Cov-2 that shares a genome sequence similar to that of other members of the beta-coronavirus group such as SARS-Cov, MERS-CoV, bat coronavirus. Our feature comprises topologicalinformation in molecular SMILES and local chemical context in the SMILES sequence for the drug chemical compound and drug target. Our proposed method prevailed with high and compelling performance accuracy of 97.30% which could beprioritized as the potential and promising prediction route for the development of novel oral antiviral medicine for COVID-19 drugs. Objective: Forecasting DTI stands as a pivotal aspect of drug discovery. The focus on computational methods in DTI prediction has intensified due to the considerable expense and time investment associated with conducting extensive in vitro and in vivo experiments. Machine learning techniques, particularly deep learning, have found broad applications in DTI prediction. We are convinced that this study could be prioritized and utilized as the promising predictive route for the development of novel oral antiviral treatments for COVID-19 and other variants of coronaviruses. Methods: This study addressed the problem of COVID-19 drugs using proposed RGCN with multifeatures as an attractive and potential route. This study focused mainly on the prediction of novel antiviral drugs against coronaviruses using graph-based methodology, namely RGCN. This research further utilized the features of both drugs and common potential drug targets found in betacoronaviruses group to deepen understanding of their underlying relation. Results: Our suggested approach prevailed with a high and convincing performance accuracy of 97.30%, which may be utilizedas a top priority to support and advance this field in the prediction and development of novel antiviral treatments against coronaviruses and their variants. Conclusion: We recursively performed experiments using the proposed method on our constructed DCCCvT graph dataset from our collected dataset with various single and multiple combinations of features and found that our model had achieved comparable best-averaged accuracy performance on T7 features followed by a combination of T7, R6, and L8. The proposed model implemented in this investigation turns out to outperform the previous related works.

The Wreaths of KHAN: Uniform Graph Feature Selection with False Discovery Rate Control

A Probabilistic Model of Functional Brain Connectivity Network for Discovering Novel Biomarkers.

Multiscale Topological Indices for the Quantitative Prediction of SARS CoV-2 Binding Affinity Change upon Mutations

Discriminative Feature Selection for Uncertain Graph Classification

Representability of algebraic topology for biomolecules in machine learning based scoring and virtual screening

Central hubs prediction for bio networks by directed hypergraph - GA with validation to COVID-19 PPI

Protein–protein and protein–nucleic acid binding site prediction via interpretable hierarchical geometric deep learning

Multiple Networks Modules Identification by a Multi-Dimensional Markov Chain Method

Graph-based Pattern Discovery in Protein Structures

A practically efficient algorithm for identifying critical control proteins in directed probabilistic biological networks

Res-GCN: Identification of protein phosphorylation sites using graph convolutional network and residual network

Relational Graph Convolution Network with Multi Features for AntiCOVID-19 Drugs Discovery using 3CLpro Potential Target

False Discovery Rate Control for Gaussian Graphical Models via Neighborhood Screening

Predicting drug-Protein interaction with deep learning framework for molecular graphs and sequences: Potential candidates against SAR-CoV-2

Optimal False Discovery Rate Control for Large Scale Multiple Testing with Auxiliary Information

Novel Graph Topology Learning for Spatio-Temporal Analysis of COVID-19 Spread.

Exploring the effectiveness of the TSR-based protein 3-D structural comparison method for protein clustering, and structural motif identification and discovery of protein kinases, hydrolases, and SARS-CoV-2's protein via the application of amino acid grouping

A novel model-free feature selection method with FDR control for omics-wide association analysis

Drug–target affinity prediction with extended graph learning-convolutional networks

Disease Gene Classification with Metagraph Representations

WDNfinder: A Method for Minimum Driver Node Set Detection and Analysis in Directed and Weighted Biological Network