CB-Smart: A Cloud-Based Machine Learning Platform for Screening Cannabinoid Receptor Ligands
- 1Wilfrid Laurier University , , , ,
- 2University of Waterloo , , , ,
Abstract
Cannabinoid receptors CB1 and CB2 are class A G protein-coupled receptors (GPCRs) implicated in pain modulation, neuroinflammation, immune regulation, and metabolic processes, making them important therapeutic targets. Conventional structure-based virtual screening for cannabinoid ligands remains computationally intensive and limited in throughput. To overcome this challenge, this work presents CB-Smart, a cloud-enabled machine learning (ML) platform to predict binding of ligands and their relative binding affinity to CB1 and CB2. Using curated experimentally validated ligands and GPCR decoy datasets, we trained five supervised ML algorithms, XGBoost, Support Vector Machines (SVM), Random Forests, Multi-Layer Perceptron (MLP), and k-Nearest Neighbors (KNN). Three featurization strategies, ECFP molecular fingerprints, RDKit 210 molecular descriptors, and Lipinski descriptors, were systematically evaluated. Models built on ECFP fingerprint features, particularly XGBoost, SVM, and MLP, demonstrated the highest performance on both reserved test datasets and an external dataset. A hard-voting ensemble integrating the top three classifiers further enhanced predictive performance. Further, SHAP-based interpretability analysis highlighted key structural and physicochemical drivers of cannabinoid receptor binding, offering insights into ligand-receptor interactions and structure-activity relationships. Finally, we integrated classification and regression models into an interactive, user-friendly cloud platform, enabling broad accessibility for virtual screening, medicinal chemistry optimization and early-stage drug discovery efforts targeting cannabinoid receptors.
Related
- The NERDD microservices platform for deploying machine learning models in drug discovery
- A machine learning-based platform for predicting the potential energy of small molecules
- Integrated machine learning, pharmacophore screening, and structure-based computational approaches for the identification of novel JAK3 inhibitors
- Machine Learning of Band Gaps and Thermodynamic Stability in A2BB′X6 (X = F, Cl, Br, I) Double Perovskites with Uncertainty-Aware Multi-Objective Screening
- AdaptDelivery: a polymer informatics platform for predicting structural properties of polymers via machine learning
BibTeX
@article{kabir2026cbsmart,
title = {{CB-Smart: A Cloud-Based Machine Learning Platform for Screening Cannabinoid Receptor Ligands}},
author = {Sahil Kabir and Amer Marwan El-Samman and Aravindhan Ganesan},
journal = {ACS Omega},
year = {2026},
month = oct,
doi = {10.1021/acsomega.6c00303},
url = {https://pubs.acs.org/acsodf/article/doi/10.1021/acsomega.6c00303/5459052/CB-Smart-A-Cloud-Based-Machine-Learning-Platform},
note = {CC BY-NC-ND 4.0},
}
RIS
TY - JOUR
TI - CB-Smart: A Cloud-Based Machine Learning Platform for Screening Cannabinoid Receptor Ligands
AU - Sahil Kabir
AU - Amer Marwan El-Samman
AU - Aravindhan Ganesan
PY - 2026
DA - 2026/10/27
JO - ACS Omega
DO - 10.1021/acsomega.6c00303
UR - https://pubs.acs.org/acsodf/article/doi/10.1021/acsomega.6c00303/5459052/CB-Smart-A-Cloud-Based-Machine-Learning-Platform
AB - Cannabinoid receptors CB1 and CB2 are class A G protein-coupled receptors (GPCRs) implicated in pain modulation, neuroinflammation, immune regulation, and metabolic processes, making them important therapeutic targets. Conventional structure-based virtual screening for cannabinoid ligands remains computationally intensive and limited in throughput. To overcome this challenge, this work presents CB-Smart, a cloud-enabled machine learning (ML) platform to predict binding of ligands and their relative binding affinity to CB1 and CB2. Using curated experimentally validated ligands and GPCR decoy datasets, we trained five supervised ML algorithms, XGBoost, Support Vector Machines (SVM), Random Forests, Multi-Layer Perceptron (MLP), and k-Nearest Neighbors (KNN). Three featurization strategies, ECFP molecular fingerprints, RDKit 210 molecular descriptors, and Lipinski descriptors, were systematically evaluated. Models built on ECFP fingerprint features, particularly XGBoost, SVM, and MLP, demonstrated the highest performance on both reserved test datasets and an external dataset. A hard-voting ensemble integrating the top three classifiers further enhanced predictive performance. Further, SHAP-based interpretability analysis highlighted key structural and physicochemical drivers of cannabinoid receptor binding, offering insights into ligand-receptor interactions and structure-activity relationships. Finally, we integrated classification and regression models into an interactive, user-friendly cloud platform, enabling broad accessibility for virtual screening, medicinal chemistry optimization and early-stage drug discovery efforts targeting cannabinoid receptors.
SN - 2470-1343
ER -