GDAS: A Machine Learning-Driven Approach for the High-Throughput Identification of Disease-Associated Glycoforms

Glycosylation is a critical post-translational modification and a rich source of disease biomarkers, yet comprehensive, site-specific identification of glycosites and glycoforms at the proteome scale remains computationally intensive and time-consuming. Here, we present Glycoproteomics Data Analysis Software (GDAS), a high-throughput computational platform that enables rapid and confident identification of disease-specific glycoforms from mass spectrometry data. GDAS integrates an ultrafast open-search strategy (e.g., MSFragger-Glyco) with statistical filtering to efficiently reduce large proteome databases to a focused set of significantly regulated glycoproteins, thereby streamlining subsequent targeted N- and O-glycosylation analyses using specialized tools, including GlycReSoft and O-Pair. A dedicated final analysis module further integrates quantitative outputs through advanced statistical modeling and machine learning algorithms, including bootstrap or Bayesian inference, as well as XGBoost and random forest classifiers, to generate robust glycosylation scores for biomarker prioritization. Application of GDAS to published Alzheimer's disease data sets demonstrates its ability to capture biologically meaningful glycosylation alterations. GDAS is freely available at https://github.com/Yang-Lab-SUMC/GDAS.