FTIR-based molecular fingerprinting for the rapid classification of dengue and chikungunya from human sera using machine learning: an observational study

Abstract

Background

Dengue and chikungunya are arboviral diseases transmitted by Aedes mosquitoes, co-endemic in southeast Asia and India. Accurate and rapid diagnosis is crucial for effective outbreak management, but conventional diagnostic methods (ELISA, RT-PCR) are limited by cross-reactivity and the need for specialized infrastructure. Vibrational spectroscopy offers a novel, label-free alternative for detecting host molecular changes directly from serum.

Methods

We conducted an observational study to evaluate the diagnostic potential of Fourier Transform Infrared (FTIR) and Raman micro-spectroscopy combined with machine learning for the classification of dengue and chikungunya from human serum. Serum samples from confirmed dengue (N = 142), chikungunya (N = 120), and healthy controls (N = 40) were analysed. Vibrational spectra were acquired using FTIR and Raman techniques, followed by spectral deconvolution and machine learning-based classification using Support Vector Machine (SVM), Neural Network (NN), and Random Forest (RF) models.

Findings

FTIR analysis revealed distinctive group-specific vibrational signatures, particularly in the Amide I and III regions, where dengue-infected sera exhibited a marked increase in β-sheet content and loss of α-helical structures. Raman spectroscopy further identified differences in nucleic acid backbone vibrations and protein conformations. The SVM, RF, and NN models, trained on FTIR data, achieved near-perfect classification (AUC = 1.000; CA-score ≥0.989), outperforming traditional diagnostic methods. Additionally, t-SNE and Silhouette analyses demonstrated superior clustering performance with FTIR, with clear separation of Chikungunya samples (average Silhouette score 0.385) compared to Raman, where clustering was less distinct.

Interpretation

Vibrational spectroscopy, particularly FTIR integrated with machine learning, offers a robust, rapid, and scalable diagnostic platform for distinguishing arboviral infections in regions with high co-infection rates. By capturing host biomolecular changes directly from serum, this method minimizes cross-reactivity and enhances diagnostic speed compared to ELISA and RT-PCR. Its deployment in point-of-care settings could significantly improve arboviral surveillance and clinical management, especially in resource-limited regions.

Funding

This study was funded by the Department of Health Research- Indian Council of Medical Research (DHR-ICMR) Grant-In-Aid grant number GIA/2020/000346 and CoEs Phase II, IIT/SRIC/IDK-PHASE-II/2024/01.