Machine Learning-Assisted Attenuated Total Reflectance Fourier-Transform Infrared Spectroscopy for Discriminating
Juthamas Jaimanee1, Patutong Chatchawal2, Molin Wongwattanakul2
1Medical Technology Program, Faculty of Associated Medical Sciences of Khon Kaen University, Khon Kaen 40002, Thailand.
Abstract:
Dengue infection remains a major global health burden, ranging from asymptomatic infection to life-threatening severe dengue, which is characterized by increased vascular permeability and leakage of plasma from the intravascular compartment into the extravascular space. Plasma leakage during the critical phase of dengue illness, which follows the febrile phase, is a major contributor to disease severity and adverse clinical outcomes. Risk stratification remains challenging because clinical warning signs and routine laboratory abnormalities may become apparent only after vascular leakage has begun. Therefore, identifying biochemical signatures associated with plasma leakage may facilitate timely risk assessment and clinical management. Herein, we employed Attenuated Total Reflectance Fourier Transform Infrared (ATR-FTIR) spectroscopy integrated with machine learning to discriminate dengue patients with and without plasma leakage during the initial febrile phase of infection. Thirty-eight confirmed dengue serum samples (15 non-plasma leakage and 23 plasma leakage) were analyzed following spectral preprocessing and machine learning classification. Group-discriminative spectra were predominantly localized to the amide I region (1700-1600 cm⁻¹), implicating alterations in protein secondary structure. Curve-fitted spectra revealed a significant increase at 1685 cm⁻¹ (β-turn-associated structure) in plasma leakage in dengue infection (p < 0.01). The absorbance ratio at 1665 cm⁻¹ to 1685 cm⁻¹ (A1665/A1685 ratio) was significantly reduced in plasma leakage relative to non-plasma leakage in dengue infection (p < 0.01). ROC analysis identified an optimal cutoff of 2.74, while logistic regression modeling yielded a sensitivity of 0.96 with an overall accuracy of 0.68. Collectively, these findings identify the A1665/A1685 ratio as a candidate spectral marker associated with plasma leakage in dengue infection and highlight the potential of ATR-FTIR based serum profiling as a rapid, label-free approach for characterizing biochemical alterations associated with the plasma leakage phenotype.
