The aim of this work was to create extended QSAR model of the relationship between sodium channel blocking activity of the particular compound and its chemical structure together with the in vitro assay conditions. Artificial neural networks (ANNs) were chosen as modeling tools. Chemoinformatics software was used for calculation of the molecular descriptors describing the structure of the interest. Drug concentration causing 50% of the channel inhibition (IC50) was used as the modeling endpoint. The data was based on the literature search and consisted of 38 drugs and 108 records. Initial number of inputs was 110 and during the sensitivity analysis was reduced to 20. ANNs models were optimized in the extended 10-fold cross-validation scheme yielding RMSE = 0.68, NRMSE = 20.7% and R2= 0.35. Best models were ANNs ensembles combining three ANNs with their outputs averaged as a collective output of the system.
keywords in English:
10-fold cross-validation, sodium channel, literature search, molecular descriptors, in-vitro assays, empirical modeling, drug concentration, chemoinformatics
affiliation:
Wydział Nauk o Zdrowiu : Instytut Zdrowia Publicznego, Wydział Farmaceutyczny : Zakład Farmacji Społecznej, Wydział Farmaceutyczny : Zakład Technologii Postaci Leku i Biofarmacji