Machine Learning Search of Novel Selective NaV1.2 and NaV1.6 Inhibitors as Potential Treatment Against Dravet Syndrome

Dravet syndrome is a type of drug-resistant and devastating childhood epilepsy, which begins in the first year of life. Etiologically, it is most frequently associated with loss-of-function de novo mutations in the gene SCN1A, which encodes for the NaV1.1 channel, a voltage-operated sodium channel highly expressed in inhibitory GABAergic interneurons. Dysfunction of this channel causes global hyperexcitability. Whereas exacerbation of seizures in Dravet patients has been observed after the administration of voltage-operated sodium channel blockers with low or no selectivity towards specific channel subtypes, recent preclinical evidence suggests that highly selective blockade of sodium channels other than NaV1.1 or the selective activation of NaV1.1 could correct the Dravet phenotype. Here, we report the development and validation of ligand-based computational models for the identification of selective NaV1.2 or NaV1.6 with no inhibitory effect on NaV1.1. The models have been jointly applied to screen the chemical library of the DrugBank 5.1.8 database, in order to select starting points for the development of specific drugs against Dravet syndrome. The ligand-based models were built using free software for molecular descriptor calculation (Mordred) in combination with in-house Python scripts. Training data was retrieved from ChemBL and specialized literature, and representatively sampled using an in- house clustering procedure (RaPCA). Linear classifiers were generated using a combination of the random subspace method (feature bagging) and forward stepwise. Later, ensemble learning was used to obtain meta-classifiers, which were validated in retrospective screening experiments before their use in the final, prospective screen.