Machine Learning-Enhanced MALDI-TOF Mass Spectrometry for Screening HBsAg-Positive Patients

Hepatitis B virus (HBV) remains a major global public health challenge, and its early screening is essential for controlling transmission and improving treatment outcomes. We analyzed serum samples from 422 participants via Matrix-assisted laser desorption/ionization time-of-flight mass spectrometry (MALDI-TOF MS) to establish a screening model for hepatitis B surface antigen (HBsAg)-positive status. Following multi-bin preprocessing and single-sample spectral aggregation, we assessed three machine learning algorithms-random forest, deep neural network, and light gradient boosting machine (LightGBM). Among them, the LightGBM model achieved the best performance, with an optimized F1 score of 0.87 and an area under the receiver operating characteristic curve (AUC) of 0.94. A 100-iteration ensemble feature stabilization strategy identified twelve distinct m/z peaks as stable biomarkers for HBsAg-positive screening. Independent validation yielded sensitivity of 77.7% and specificity of 76.0%-insufficient for individual diagnosis but potentially suitable for population-level surveillance programs combined with confirmatory testing, particularly in resource-limited settings where conventional methods are impractical. Notably, the method offers a detection time of approximately one minute, a per-sample cost of ~$0.14. In conclusion, the combination of MALDI-TOF MS and machine learning enables a rapid, low-cost screening tool for large-scale HBV detection.