Tumour marker analysis using a machine learning assisted vibrational spectroscopy approach

Tumour biomarkers such as CA125, CA15-3, CA19-9, AFP and CEA are routinely used in the oncology clinic to diagnose cancer, monitor response to therapy, and detect relapse. However, their quantification depends on immunoassay-based methods that are time-consuming, reagent-dependent, and poorly suited to resource-limited settings. Here, we present a machine learning-assisted ATR-FTIR spectroscopy approach for label-free tumour biomarker analysis to enable simple and rapid quantification at the bedside. Using principal component analysis (PCA), we first demonstrate that these five clinically relevant biomarkers are spectrally separable, with the protein-associated region (1200-1700 cm -1 ) providing the greatest discriminative information. We then develop partial least squares regression (PLSR) models to quantify CA125 in phosphate-buffered saline (R 2 = 0.95) and in human serum across a clinically relevant concentration range, achieving reliable predictions at and above the clinical decision threshold of 35 U/mL. A semi-quantitative classification model further demonstrated robust identification of elevated CA125, with a macro-average sensitivity of 0.86 and specificity of 0.92. These results support ATR-FTIR spectroscopy as a rapid, reagent-free platform for cancer biomarker monitoring, with potential utility in resource-limited settings.