ML-based Models as a Strategy to Discover Novel Antiepileptic Drugs Targeting Sodium Receptor Channel

Area of Science:

• Computational chemistry and cheminformatics
• Pharmacology and drug discovery
• Machine learning in bioinformatics

Background:

• Epilepsy is a common chronic neurological disorder requiring long-term management.
• Current treatments aim to manage symptoms, but novel therapeutic strategies are needed.
• Voltage-gated sodium channels (VGSCs) are implicated in epilepsy due to increased neuronal excitability.

Purpose of the Study:

• To compare machine learning classifiers for predicting the inhibitory activity of compounds against the human NaV1.7 protein.
• To identify the most effective machine learning models for classifying molecules as active or inactive against this epilepsy target.
• To leverage computational tools for accelerating the discovery of potential epilepsy therapeutics.

Main Methods:

• Utilized Weka software for machine learning analysis on a dataset of 1781 compounds from ChEMBL.
• Computed molecular fingerprints using the ChemDes server for quantitative structure-activity relationship (QSAR) analysis.
• Evaluated various Weka classifiers (e.g., OneR, J48, Bagging) using metrics like accuracy, RMSE, ROC, MCC, precision, recall, and F-measure.

Main Results:

• The OneR classifier demonstrated superior performance in predicting active, inactive, and intermediate compounds against the NaV1.7 protein.
• J48 and Bagging classifiers also exhibited high accuracy, achieving an MCC of 1, ROC area of 1, and near-zero RMSE.
• These results highlight the effectiveness of ML models in classifying potential drug candidates.

Conclusions:

• Machine learning tools offer a rapid, cost-effective method for identifying potential drug inhibitors.
• OneR, J48, and Bagging models effectively distinguish active and inactive compound classes targeting the human NaV1.7 protein.
• These predictive models can significantly aid in the rational design of novel anti-epileptic drugs.