Drug repurposing for hyperlipidemia associated disorders: An integrative network biology and machine learning approach. (June 2021)

• Record Type: Journal Article
• Authors: Rai, Sneha; Bhatia, Venugopal; Bhatnagar, Sonika
• Is Part Of: Computational biology and chemistry. Volume 92(2021)
• Publication Date: 2021-06
• Subjects: CVD Cardiovascular diseases -- DTN drug-target network -- DDTN directed drug-target network -- HL hyperlipidemia -- HBC high betweenness centrality -- HDL-c high density lipoprotein cholesterol -- LDL-c low density lipoprotein cholesterol -- MDS minimum driver node set -- PI3K phosphatidylinositol-4, 5-bisphosphate 3-kinase; Repurposing -- Drug-target network -- Hyperlipidemia -- Driver nodes -- Machine learning; Chemistry -- Data processing -- Periodicals; Biology -- Data processing -- Periodicals; Biochemistry -- Data processing; Biology -- Data processing; Molecular biology -- Data processing; Periodicals; Electronic journals; 542.85
• DOI: 10.1016/j.compbiolchem.2021.107505
• Languages: English
• ISSNs: 1476-9271

Abstract

Graphical abstract: Highlights: The study uses a hybrid method combining network biology and machine learning. Thirty-four signaling pathways were used to construct a directed protein-protein interaction network. The driver nodes of hyperlipidemia were found associated with other diseases. A Random forest classifier was trained on 1445 molecular descriptors. Nine repurposing drug candidates have been proposed for hyperlipidemia and its associated diseases. Abstract: Hyperlipidemia causes diseases like cardiovascular disease, cancer, Type II Diabetes and Alzheimer's disease. Drugs that specifically target HL associated diseases are required for treatment. 34 KEGG pathways targeted by lipid lowering drugs were used to construct a directed protein-protein interaction network and driver nodes were determined using CytoCtrlAnalyser plugin of Cytoscape 3.6. The involvement of driver nodes of HL in other diseases was verified using GWAS. The central nodes of the network and 34 overrepresented pathways had a critical role in Hyperlipidemia. The PI3K-AKT signalling pathway, non-essentiality, non-centrality and approved drug target status were the predominant features of the driver nodes. Next, a Random Forest classifier was trained on 1445 molecular descriptors calculated using PaDEL for 50 approved lipid lowering and 84 lipid raising drugs as the positive and negative training set respectively. The classifier showed average accuracy of 76.8 % during 5-fold cross validation with AUC of 0.79 ± 0.06 for the ROC curve. The classifier was applied to select molecules with favourable properties for lipid lowering from the 130 approved drugs interacting with the identified driver nodes. We have integrated diverse network data and machine learning to predict repurposing of nine drugs for treatment of HL associated diseases.