Leveraging a Meta-Learning Strategy to Advance the Accuracy of Neutralizing Antibodies against Dengue Virus Serotype Prediction
Issued Date
2025-12-23
Resource Type
eISSN
24701343
Scopus ID
2-s2.0-105025656304
Journal Title
ACS Omega
Volume
10
Issue
50
Start Page
61296
End Page
61307
Rights Holder(s)
SCOPUS
Bibliographic Citation
ACS Omega Vol.10 No.50 (2025) , 61296-61307
Suggested Citation
Charoenkwan P., Pipattanaboon C., Schaduangrat N., Mahmud S.M.H., Shoombuatong W. Leveraging a Meta-Learning Strategy to Advance the Accuracy of Neutralizing Antibodies against Dengue Virus Serotype Prediction. ACS Omega Vol.10 No.50 (2025) , 61296-61307. 61307. doi:10.1021/acsomega.5c05741 Retrieved from: https://repository.li.mahidol.ac.th/handle/123456789/113743
Title
Leveraging a Meta-Learning Strategy to Advance the Accuracy of Neutralizing Antibodies against Dengue Virus Serotype Prediction
Corresponding Author(s)
Other Contributor(s)
Abstract
Dengue remains a major global public health threat with millions of infections reported annually and no widely effective treatment options available. Therapeutic monoclonal antibodies, particularly broadly neutralizing antibodies (bNAbs), show promise as treatments due to their ability to target conserved viral epitopes across all DENV serotypes. However, characterizing dengue virus (DENV) through experimental approaches remains expensive and time-consuming. Therefore, computational methods capable of identifying bNAbs against DENV from antibodies (CDR-H3) and epitope information can greatly complement experimental approaches and facilitate the rapid screening of bNAbs. Here, we propose an innovative meta-learning approach, termed Meta-iNAb, that can accurately identify bNAbs against DENV based on CDR-H3 and epitope information. In Meta-iNAb, to extract the diverse information on bNAbs, we employed 14 different feature encoding methods, incorporating sequential information, physicochemical properties, and composition–transition–distribution information. Then, each feature descriptor was used to construct base-classifiers using 12 popular machine learning (ML) algorithms. Finally, the 9 informative base-classifiers were selected using our customized genetic algorithm and subsequently combined to construct our meta-classifier. Benchmarking experiments revealed that Meta-iNAb is highly effective, outperforming the existing method and its base-classifiers during independent testing, with an accuracy of 0.851, an MCC of 0.702, an F1 score of 0.836, and an AUC of 0.883. To enable the rapid and efficient identification of bNAbs against DENV, an online web server for Meta-iNAb (https://pmlabqsar.pythonanywhere.com/Meta-iNAb) has been implemented. This innovative method is anticipated to serve as a high-accuracy and efficient tool for the rapid screening of stable and potent NAbs targeting DENV-1 to DENV-4.