Abstract
This study aimed to investigate the potential of peptide mass fingerprints (PMFs) of the serum peptidome using matrix-assisted laser desorption/ionization time-of-flight mass spectrometry (MALDI-TOF MS), in combination with machine learning algorithms-support vector machine (SVM) and random forest (RF)-for the diagnosis and classification of hepato-pancreato-biliary (HPB) cancers. Serum samples collected from healthy individuals and patients with various HPB cancers were analyzed to generate PMF profiles. The resulting data were randomly split into training and testing sets. Feature selection on the training set identified 71 informative peptide mass fingerprints, which were then used to construct predictive models using SVM and RF algorithms. Visualization using heatmap, PLS-DA, and multiclass RF analysis showed clear separation between healthy individuals and HPB cancer patients, as well as among different HPB cancer subtypes. Both models achieved high classification performance, with accuracy, AUROC, and MCC values exceeding 0.90 in both training and testing datasets. Notably, the models also exhibited strong multiclass discrimination ability. These findings demonstrate that serum PMF profiling using MALDI-TOF MS, combined with SVM and RF models, enables high-performance, non-invasive detection and classification of HPB cancers, with strong potential to support early diagnosis and inform clinical decision-making.