ATR-FTIR Spectroscopy Combined with Machine Learning Enables Detection of Schistosomiasis-Associated Biochemical Signatures in Human Serum

Abstract Schistosomiasis, caused by Schistosoma mansoni, remains a major public health burden in endemic regions, demanding rapid and reliable diagnostic strategies. Here, we report a label-free diagnostic platform based on attenuated total reflectance Fourier transform infrared (ATR-FTIR) spectroscopy combined with machine learning for the detection of infection-associated biochemical signatures in human serum. A total of 112 serum samples (54 infected and 58 noninfected individuals) were analyzed, with parasitological examination revealing only S. mansoni eggs in all positive samples, with no other parasitic coinfections detected. Spectral data were preprocessed and classified using supervised machine learning algorithms. First-derivative preprocessing combined with a Random Forest classifier showed the best performance in the initial cross-validation analysis, achieving 0.80 accuracy, 0.81 sensitivity, 0.79 specificity, an area under the curve (AUC) of 0.83, and an F1-score of 0.80. To further assess model robustness, nested cross-validation yielded an accuracy of 0.76 ± 0.07, sensitivity of 0.76 ± 0.16, specificity of 0.77 ± 0.19, AUC of 0.83 ± 0.11, and F1-score of 0.76 ± 0.08, providing a more conservative estimate of performance. SHapley Additive exPlanations (SHAP) analysis identified spectral regions associated with proteins, nucleic acids, and carbohydrates that contributed to model predictions. Additional sensitivity analyses indicated spectral differences between endemic-area and nonendemic-area controls and reduced performance in cross-site evaluation, highlighting geographic and cohort-related variation as a potential limitation to model generalizability. Overall, ATR-FTIR spectroscopy combined with machine learning shows potential as a rapid, reagent-free, and minimally sample-intensive complementary approach for schistosomiasis screening, warranting validation in larger, geographically diverse cohorts.