Molecular Modeling and Analysis of Cannabinoid and Cannabinoid-like Molecules Combining K-Means Clustering with Pearson Correlation and PCA

More recently, cannabinoid molecules have been widely studied for their potential to treat various diseases. We used a multidisciplinary approach, combining molecular docking and machine learning tools, to identify cannabinoid-based molecules as potential acetylcholinesterase inhibitors. We brought together molecules from the classes of cannabinoids, stilbenoids, isoflavones, and other natural products, along with their electronic structure and absorption, distribution, metabolism, excretion and tolerable toxicity (ADMET) data. A novel machine learning framework (MolSimEx, Molecular Similarity Explorer) combining K-means clustering, Pearson correlation, and principal component analysis was developed to address the similarities of these groups. From the dataset, 30 molecules were selected based on docking scores below -11 kcal/mol. The K-means clustering yielded high classification accuracy on the dataset, correctly grouping the cannabinoid analogues. Additionally, these analogues clustered with classical acetylcholinesterase inhibitors such as huprine-X, huprine-W, and donepezil when considering ADMET and electronic descriptor data. Radulanin J showed the highest correlation (0.41) with donepezil's profile, suggesting the potential of cannabinoid-derived compounds as acetylcholinesterase inhibitors.