Desain Molekul Kandidat Inhibitor Monoamine Oxidase-B Menggunakan Model Drugex Dan Evaluasi Admet Secara in Silico

Abstract

Parkinson’s disease is a neurodegenerative disorder characterized by the degeneration of dopaminergic neurons, leading to a decrease in dopamine levels in the brain. One widely developed therapeutic strategy is to inhibit the enzyme monoamine oxidase B (MAO-B) to prevent the degradation of dopamine. This study aims to design de novo candidate MAO-B inhibitor molecules using the reinforcement learning-based generative model DrugEx, predict their biological activity using a Quantitative Structure–Property Relationship (QSPR) model, and evaluate their drug-likeness properties and ADMET profiles in silico. The MAO-B inhibitor dataset was obtained from the ChEMBL database and used to build a Random Forest-based QSPR model as the reward function in the reinforcement learning process. The results show that the generative model successfully created 1,000,000 new molecules, with 60,316 molecules having a predicted pIC50 value ≥ 6.5 and 911 molecules having a predicted pIC50 value > 7.5. Selection of the 1,000 molecules with the highest predicted pIC50 values yielded 959 molecules with a SAScore ≤ 4.5 and 554 molecules that met all Lipinski’s Rules criteria. Further evaluation of ADMET profiles, structural similarity analysis, and novelty analysis yielded 49 candidate MAO-B inhibitor molecules with favorable ADMET profiles, low structural similarity to approved MAO-B inhibitors (with a maximum Tanimoto coefficient of approximately 0.34), and a high degree of novelty based on a search of the Chemspace database. The integration of the DrugEx generative model, QSPR predictions, and ADMET evaluation proved effective in identifying new MAO-B inhibitor candidates with potential for further development through molecular docking studies, molecular dynamics simulations, and experimental validation.