NATPRO Journal

Rotational energy barrier of cannabidiol by dynamic NMR

Received: September 11, 2023; Revised: November 28, 2024; Accepted: November 30, 2023; Published: December 3, 2023

NATPRO J (2023) 1: 1-4 Article

Authors: Jaehong Han*, Aida Bibi, Huynh Thi Ngoc Mi, and Heji Kim

Metalloenzyme Research Group and Department of Plant Science and Technology, Chung-Ang University, 4726 Seodong-daero, Anseong 17546, Republic of Korea 

https://doi.org/10.23177/NJ023.901

Abstract: Cannabidiol, CBD, as a non-narcotic substance, has drawn a great research interest due to the clinical and pharmaceutical significance for the treatment of pain, inflammation, anxiety, and insomnia. To provide the basic molecular dynamics of CBD, the rotational energy barrier of the aromatic group in CBD has been studied by means of temperature-dependent dynamic NMR spectroscopy and DFT calculations. Among the four stable CBD conformers, two rotamers were found as major contributors of the solution NMR structure of CBD by DFT/B3LYP/6-311+G(2d,p)-level calculation. Between two, the rotational rate constant k = 175 s-1 at 35°C and rotational energy barrier ΔG‡=14.5 kcal/mol were determined by Eyring equation. The result will provide a fundamental information for the sophisticated ligand-receptor docking study between CBD and biological receptors at the molecular level.

Keywords: CBD, DFT, dynamic NMR, rotational energy, temperature-dependent 

Deep Learning-Based Activity Prediction for Estrogen Receptor Modulators and Application to Natural Product Screening 

Received: September 11, 2024; Revised: November 20, 2024; Accepted: February 11, 2025; Published: February 14, 2025

NATPRO J (2024) 1: 5-15 Article

Authors: Jongkeun Choi*

Department of Biological and Chemical Engineering, Chungwoon University, Sukgol-ro 113, Michuhol-gu Incheon, 22100, Republic of Korea

https://doi.org/10.23177/NJ024.901

Abstract: This study investigated the use of deep learning techniques to predict estrogen receptor (ER) activity in small molecules, crucial for drug discovery in hormone-dependent diseases. Active and inactive compound data were collected from PubChem and BindingDB, and their chemical properties, including molecular weight, polar surface area, and hydrogen bond characteristics were analyzed. Utilizing RDKit, molecular descriptors and Morgan Fingerprints were calculated. Chemical space analysis using principal component analysis (PCA) visualization revealed that this approach alone was insufficient for distinguishing between active and inactive compounds. Therefore, two TensorFlow-based deep learning models were developed: one using molecular descriptors and the other using Morgan Fingerprints. Both models were trained on BindingDB and PubChem datasets with varying activity thresholds and molecular weight restrictions. The Morgan Fingerprint-based model consistently outperformed, achieving up to 99.82% accuracy and 0.9994 AUC on the BindingDB dataset. To validate practical applicability, 81,442 compounds from the NPASS natural product database were screened using the best-performing model. This virtual screening identified 3,577 potential ER-active candidates, including known active compounds and novel potential modulators. The results highlighted the superiority of Morgan Fingerprints in capturing relevant structural features for activity prediction and emphasized the importance of high-quality datasets in model development. This study also demonstrated the potential of deep learning in expediting drug discovery processes, particularly in identifying promising candidates from large compound libraries. Future work will need to include free energy calculations using molecular dynamics and experimental validation. 

Keywords: Artificial intelligence, deep learning, estrogen receptor, drug discovery, TensorFlow, ligand activity prediction, Morgan Fingerprints, virtual screening