Please use this identifier to cite or link to this item:
Title: Identification of Novel Inhibitors and Biomarkers for Alzheimer's Disease using Computational Systems Polypharmacology and Machine Learning Approach
Authors: Shukla, Rohit
Singh, Tiratha Raj [Guided by]
Keywords: Novel Inhibitors
Alzheimer's Disease
Computational Systems
Machine Learning
Issue Date: 2022
Publisher: Jaypee University of Information Technology, Solan, H.P.
Abstract: The current research is focused on the management of the pathophysiology of the Alzheimer’s Disease (AD). The AD is a neurological irreversible disease characterized by the abnormal accumulation of amyloid beta and neurofibrillary tangles in the brain. Although these hallmarks can identify in the later stages. Currently used drugs for AD can only slowdown the progression but they cannot halt the AD progression. We have analysed thousands of AD genes and proposed the biomarker identification method as well as > 1 million compounds were screened against several potential targets followed by 5 µS molecular dynamics simulation (MDS) and lead compounds were proposed. Firstly, we have retrieved the AD genes from the DisGeNet dataset and 13,504 features were calculated. These features were evaluated by using 16 machine learning (ML) methods. The result showed that network-based features are showing ∼92% accuracy while sequence-based features only showing ∼52% accuracy. The feature selection approach increases ∼2-3% accuracy. Best performing features were used for the feature fusion analysis. By utilizing feature fusion approach, we have constructed 24 new features with 6,020 dimensions. The composition of k spaced amino acid pairs (CKSAAP) based fused features showed ∼10% increase in the accuracy. Then 8 CKSAAP fused features with alone CKSAAP were considered for hyperparameter tuning where we have not seen >=70% accuracy for any feature. Therefore, we left the sequence-based features and used network-based features for the hyperparameter tuning approach where we have seen that network-based features are able to classify between the AD and non-AD genes with 97.23 and 96.55% accuracy for training and test dataset respectively. Then proposed model was validated by using the blind dataset obtained from AlzGene and Gene Expression Omnibus databases. Finally, we have proposed AlzGenPred method as a standalone tool to the scientific community. The GSK3β and CDK5 are two key enzymes which majorly phosphorylate the tau protein. We have taken these two targets and screened natural and drug like compounds (> 1 million compounds) and selected potential compounds for ADMET analysis. Based on ADMET evaluation, we have selected 20 and 17 compounds from natural compounds dataset and 29 and 16 compounds from drug-like compounds dataset which are the best fit in all these parameters. MDS of 100 ns were carried out and then several structural parameters and binding free energy were calculated for lead selection which revealed that for GSK3β (ZINC15968620, ZINC15968622, ZINC70707119 from natural compound dataset and ZINC21011059, ZINC21011066 from drug-like compounds dataset) and CDK5(ZINC85877721, ZINC96116231 from natural compound dataset and ZINC6261568, ZINC14168360 from drug-like compounds dataset) can act as a potential inhibitor to reduce the tau phosphorylation. AD is a multifactorial disease where multiple targets simultaneously activate and contribute in the disease progression therefore single target inhibition cannot halt the AD progression. Thus, we have designed an AD virtual cell where we have observed the effect of multitargeted drug against several target. The information of genes, proteins and metabolites were collected from the experimental findings and then an AD virtual cell was constructed by using the Cell Designer 4.0. After that we have increased the drug concentration till 3.0 at 0.5 interval and we have observed that the drug dose increment is decreasing the neuronal death and increasing the neuronal survival. Then we have analysed the effect of drug against individual target where we have seen that inhibiting single target is not instantly reducing the plaques and tangles amount but inhibition of all the targets are instantly reducing the plaques and tangles burden. Increasing the concentration of all the targets were highly participating in the disease progression followed by neuronal death increase while single target concentration increment is not increasing the plaques and tangles in high amount. Therefore, from this finding we have concluded that the multitargeted drug is more beneficial than single target drug. Hence, we have taken 1,416 natural compounds and screened against the selected eight targets. Then based on the >= -7.0 kcal/mol binding energy cut-off, we have selected 248 potent compounds for pharamacokinetics evaluation. It reveals that two compounds are good in all these parameters therefore 32 systems were created and employed for MDS. Based on the simulation result we have found that these two ligands (ZINC898396 and ZINC14586979) are stable against the binding pocket of respective proteins. These selected ligands were inputted in to the SwissTarget database for additional AD target identification. We have taken 100 targets for each ligand and then those targets were matched with the AD genes. The PPI network were generated and analysed where we have not found any additional target for the ZINC898396 while two additional targets Caspase-3 (CASP-3) and Nitric Oxide Synthase-3 (NOS-3) for ZINC14586979 were predicted. ZINC14586979 were docked against CASP-3 and NOS-3 followed by 100 ns MDS. Docking and MDS analysis clearly revealed that ZINC14586979 is stable towards the binding cavity of these two targets. Therefore, finally we have proposed that ZINC898396 (Annonine) can inhibit eight targets while ZINC14586979 (Marcanine A) can inhibit ten targets and can act as multitargeted directed ligands. These compounds can be further tested by using in-vitro and in-vivo experiments.
Description: PHD0251, Enrollment No. 186501
Appears in Collections:Ph.D. Theses

Files in This Item:
File Description SizeFormat 
PHD0251_ROHIT SHUKLA_186501_BI_2022.pdf20.76 MBAdobe PDFView/Open

Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.