Ligand-based virtual screening of a benzylisoquinoline alkaloids dataset with anti-inflammatory potential activity.

Inhibitor of nuclear factor kappa B kinase beta subunit (IKK-B) and extracellular signal-regulated kinase 1 (IKK-B) are two proteins involved in cytokine intracellular signaling pathways, which have a great importance due to their anti-inflammatory role. In this work, from the ChemBL database were obtained 775 and 48 structures with activity against IKK-B (CHEMBL1991) and ERK1 (CHEMBL3385) respectively. The compounds were classified using values of pIC_50, presenting a range of 4.29 (from 5.01 to 9.30) for IKK-B and 3.10 (From 5.05 to 8.15) for ERK-1. From SMILES codes, 2D structures were generated in Standardizer and after calculated 1064 two-dimensional molecular descriptors in Dragon 7 software. Obtained results were imported to Knime 3.1.0 software. All variables were submitted to autoscaling and after were partitioned to generate two groups, a training group composed by the 80% of the whole molecules set and a test group composed by the remaining 20%. (Q)SAR models was performed using a Random Forest algorithm.  Models were evaluated through cross validation (leave-one-out), Q ²_LOO = 0.69 and 0.66 as well as external test, Q ²_ext = 0.74 and 0.58 for IKK-B and ERK1 respectively. Finally, pIC50 value of 179 benzylisoquinoline alkaloids were predicted in the (Q)SAR models found 4 compound with the highest activity for each one protein studied.