What question did this study set out to answer?

The research aims to leverage machine learning to discover inhibitors targeting the Marburg virus Gene 4 Small ORF protein, essential for viral replication.

February 8, 2026BMC Infectious Diseases0 citationsOpen Access

MARVpred: machine learning prediction of inhibitors targeting Marburg virus Gene 4 Small ORF protein

ELEugene LampteyUniversity of Ghana GAGabriel AnyaeleUniversity of Ghana HAHarry E. ArthurUniversity of Ghana

Key Points

The research aims to leverage machine learning to discover inhibitors targeting the Marburg virus Gene 4 Small ORF protein, essential for viral replication.
Used a dataset from PubChem consisting of 301,745 compounds.
Trained various machine learning models, including Random Forest and Gradient Boosting Machines.
Classified activity outcomes as active or inactive using molecular descriptors.
Employed an 80:20 split for training and validation of models with performance metrics like accuracy and AUC-ROC.
Morgan fingerprints achieved higher predictive accuracy (76%), precision (80%), and ROC-AUC (84%) compared to PubChem fingerprints.
Random Forest and Gradient Boosting Machines were the top-performing models.
Random Forest attained the highest specificity (83%) and ROC-AUC (0.84).
Validation on new datasets confirmed strong predictive reliability of the models.

Abstract

The Marburg virus (MARV), responsible for severe hemorrhagic fevers with mortality rates as high as 90%, remains a significant public health threat. This study employs machine learning to identify inhibitors targeting the MARV Gene 4 Small ORF protein, crucial for the virus's replication and immune evasion. The Gene 4 Small ORF protein is pivotal in taking over the host's cellular mechanisms, facilitating unchecked viral replication and significant immune system disruption. Effective targeting of this protein holds promise for mitigating the viral lifecycle and entry, potentially curbing the severity of the disease outbreaks. A dataset from PubChem, including 301,745 compounds, was utilized to train models like Random Forest (RF), Gradient Boosting Machines (GBM), CatBoost (CB), AdaBoost (AB), and Logistic Regression (LR). The activity outcomes were classified with integers active as 1 and inactive as 0, followed by molecular descriptor generation using RDKit and PaDEL. The models were trained on an 80:20 split and validated on a novel dataset to ensure robustness, with performance metrics such as accuracy and AUC-ROC guiding evaluation. Morgan fingerprints outperformed PubChem fingerprints, achieving higher accuracy (76%), precision (80%), and ROC-AUC (84%). Among the machine learning models evaluated, RF and GBM were the best performers, with RF achieving the highest specificity (83%) and ROC-AUC (0.84). Validation on new datasets further confirmed the effectiveness of these models, with RF and GBM demonstrating strong predictive reliability for identifying potential inhibitors of the Marburg virus. A Web Application known as MARVpred was developed to predict the activity of compounds with anti-MARV properties from the ChEMBL database. MARVpred is freely accessible online (https://igmr.org/software/marvpred). This study signifies a critical step forward in the computational prediction of viral inhibitors, offering a valuable tool for accelerating the development of Marburg virus therapeutics.

Demander à l'IA

Bookmark

View Full Paper