Improved analytical workflow towards machine learning supported N-glycomics-based biomarker discovery

The composition and function of glycans are very complex thus manual data interpretation of their structural elucidation is difficult. Capillary electrophoresis is one of the liquid phase separation techniques, which is most frequently used to address these challenging tasks. Combining high-resolution capillary electrophoresis with machine learning-supported data interpretation holds the promise to gain as much chemical and clinical information from the analyzed samples as possible. However, this combination requires significant technological improvements both in the analytical and the data processing aspects. In this study we report on the development of an automated, liquid-handling robot-based sample preparation method to obtain reproducible and N-glycome profiles by capillary electrophoresis for the subsequent machine learning-supported data interpretation, which was optimized for the special needs of the analysis. The resulting new glycoanalytical workflow was then tested for a demanding problem to predict the effectiveness of chemotherapy treatments of lung cancer patients ensuring the effective management of the disease. Our findings revealed that the achieved N-glycan data contained important clinical information to accurately predict patient response to chemotherapy with AUC values ranged from 0.8290 to 0.8410.