Multiscale Computational Pipeline for Anti-Dengue Drug Discovery Using AI, Docking, and Network Pharmacology

Finding potential antiviral inhibitors has advanced significantly through computer-aided drug discovery.

The Random Forest regression (RFR), PLS regression, NNKernel regression, and pcRegression models were compared and assessed using R². In the following step, molecular docking was conducted to assess binding affinities and protein-ligand interactions. The ligand with the highest ICM score was subjected to molecular dynamics simulations to analyze binding and mechanism. In the final step, network pharmacology was adopted to study the mechanism of action.

The R² value for RFR is 0.93, indicating that the model performed exceptionally well in terms of prediction, with highly correlated experimental and predicted values. The ML model showed that only 152 out of the 186 compounds have pIC50 values greater than 6. Ligands with pIC50 values greater than 6 were selected for molecular docking with the NS2B/NS3 protease. Molecular docking studies revealed that 28 of the 152 compounds have ICM scores below 0, and five compounds were selected at the top of the list based on their binding interactions and ICM scores. The compound that was ranked highest, PubChem IDs 44626355, 166633044, 166633428, 20873605, and 166626117, had ICM scores of -21.657148, -19.151558, -16.963572, -16.836168, and -15.635322 kcal/mol.

PubChem CID: 44626355 compound's binding stability is attributed to strong hydrogen bonding, van der Waals, and hydrophobic interactions. Molecular dynamics simulations, PCA, and FEL were conducted. Through network pharmacology, the therapeutic mechanism emerges that acts across several interconnected pathways involved in dengue pathology.

Our study offers a promising framework for identifying potent inhibitors of the dengue virus.