CandyCrunch is a tool that predicts glycan structures using machine learning based on fragmentation patterns and propensities in MS/MS data. The tool was trained on a large dataset of annotated LC-MS/MS spectra derived from various glycan classes, providing a representative view of current glycomics data. CandyCrunch uses a dilated residual neural network architecture suited for MS data and considers experimental parameters to predict glycan rankings accurately.
Furthermore, CandyCrunch is complemented by CandyCrumbs, a Python-based solution for automating the annotation of fragment ions in glycan structures. CandyCrumbs can annotate fragment ions in Domon-Costello and IUPAC-condensed nomenclature rapidly and comprehensively, providing a feature-complete implementation for this task.
The tool was validated through molecular dynamics simulations, revealing insights into the fragmentation mechanisms of specific glycan isomers. CandyCrunch and CandyCrumbs were then applied to derive new biological insights from glycomics data, such as predicting the glycomes of different cell lines and identifying differential glycan expression patterns between them.
Overall, CandyCrunch and CandyCrumbs offer a powerful platform for high-throughput glycomics analysis, enabling researchers to save time, enhance annotation robustness, and derive new biological insights from glycan data. The tools can be used to predict glycan structures accurately, automate fragment ion annotation, and explore complex glycan networks to uncover novel findings in glycomics research.
Source link
Source link: https://www.nature.com/articles/s41592-024-02314-6