FTIR spectroscopy of peripheral blood mononuclear cells and machine learning: Spectral biomarkers for bacteremia, focal bacterial, and viral infections.

The rapid and accurate differentiation of bacteremia, focal bacterial infections (FBI), and viral illnesses in febrile immunocompromised patients is a critical unmet clinical need that directly impacts antibiotic stewardship. While host-response diagnostics are increasingly applied in clinical practice, most current approaches cannot reliably differentiate systemic from localized bacterial infections and do not offer mechanistic, biochemically interpretable signatures underlying their classifications. This study introduces a unified platform for Fourier-transform infrared (FTIR) spectroscopy and machine learning that balances diagnostic performance and biochemical explainability. Using white blood cells from 457 pediatric oncology episodes (69 bacteremia, 90 FBI, 178 viral, and 120 control), the platform was evaluated on key diagnostic challenges. It achieved high accuracy in distinguishing bacteremia from FBI (94.6%) and bacteremia from viral infections (93.5%). Crucially, it also demonstrated robust performance in the critical comparison of FBI versus viral infections, a common diagnostic dilemma. To elucidate the biochemical basis of this performance, we identified interpretable spectral biomarkers for each etiology. Bacteremia exhibited systemic inflammatory signatures in carbohydrate/phosphodiester regions (∼1160 cm-1), while FBI showed features in amide II and lipid regions (∼1580-1560, ∼1430 cm-1). This work establishes a diagnostic platform that successfully resolves fine-grained infection etiologies and directly links its high accuracy to underlying, explainable immunometabolic host responses, offering a powerful tool for combating antimicrobial resistance.