Note for: HCVpred: A web server for predicting the bioactivity of hepatitis C virus NS5B inhibitors

Note for: HCVpred: A web server for predicting the bioactivity of hepatitis C virus NS5B inhibitors

Doi: 10.1002/jcc.26223

Overview

HCV

-          NS5B -- > one of three potential therapeutic target HCV (other two – NS3/4A, NS5A) -- > involve in viral replication

-          Developing classification structure-activity relationship (CSAR) model -- > for identifying substructures giving rise to anti-HCV from 578 non-redundant compounds

-          NS5B inhibitors -- >

o   Set of 12 fingerprint descriptors -- > use as predictive models

o   Random forest algorithm with 100 independent data splits

-          Modelability (MODI index) of data se

o   Robust

o   0.88 exceeding – threshold 0.6

-          Predictive performance

o   Accuracy

o   Sensitivity

o   Specificity

o   Matthews coefficient

-          In-depth analysis of top 20 important descriptors

o   Aromatic ring and alkyl side chains -- > important for NS5B inhibition

-          Predictive model is publicly deposited -- > allow user to predict biological activity

o   http://codes.bio/hcvpred/

-          Web-server -- > use for designing more potent and specific drugs against HCV NS5B

Extract the flow

-          Structural + chemical functions of compounds/inhibitors and biological activity

o   Understand the relationship between physicochemical properties on biological activity

o   Physicochemical properties of compound -- > computed through molecular descriptors calculation software -- > obtain both quantitative value + qualitative label

-          Data compilation and curation

o   HCV NS5B -- > ChEMBL ID: CHEMBL5375

o   1903 bioactivity data points from 1350 compounds

§  IC50 is used as bioactivity unit for further investigate

§  865 compounds

§  Define threshold

·         <1 uM active

·         > 10 uM inactive

·         Intermediate is excluded (1-10 uM) -- > 284 inhibitors

·         Final non-redundant and curated compounds -- > 578 inhibitors

-          Fingerprint descriptors extraction

o   PaDEL-descriptor -- > get 12 molecular fingerprints from 9 classes

§  AtomPairs 2D,

§  CDK fingerprinter,

§  CDK extended,

§  CDK graph only,

§  E-state,

§  Klekota–Roth,

§  MACCS,

§  PubChem,

§  Substructure

-          415 – active compounds, 163 – inactive compounds

-          Data set modelability

o   Similar chemical structure

§  Give rise to different bioactivity -- > not good for the modeling

o   Similar chemical structure

§  Give rise to similar bioactivity -- > good for the modeling

o   MODI -- > modelability index -- > telling you whether the model is good or bad -- > this study set MODI -- > >0.65 -- > considered to be modelable

-          Webserver

o   Using shinny app

§  User interface (ui.R)

·         Input data (SMILE string)

·         PaDEL-descritor -- > convert SMILE to chemical fingerprint

·         Analyze through predictive model

§  Server interface (server.R)

-          Input -- > SMILES string

o   Passing through the descriptor extraction (PaDEL)

o   Go through RF machine learning

o   Classification labeling the input SMILES, active or inactive -- > output format .csv