O-GlyThr

O-GlyThr predicts human O-linked threonine glycosylation sites using a feature-fusion approach and a random forest classifier to enable accurate identification of O-glycosites.

Key Features:

• Target residue and organism: Predicts O-linked glycosylation sites on threonine residues in Homo sapiens proteins.
• Feature encoding: Integrates seven distinct feature coding methods to represent sample sequences comprehensively.
• Classifier: Uses a random forest classifier selected after evaluation of various algorithms.
• Training data: Built from collected and curated high-quality human protein data with known O-linked threonine glycosylation sites.
• Validation and evaluation: Validated by rigorous 5-fold cross-validation and assessment on an independent validation dataset.
• Performance metrics: Achieved AUC 0.9308 on the training set and AUC 0.9323 on an independent validation dataset, and ACC 0.8475 on an independent test dataset.
• Comparative benchmarking: Compared against previously published predictors, achieving the highest reported ACC of 0.8475 on an independent test dataset.

Scientific Applications:

• Glycosite identification: Enables identification of O-glycosites on threonine residues in human proteins for glycobiology research.
• Glycosylation structure–function studies: Supports studies of glycosylation structure and function in cellular metabolic and signaling pathways.

Methodology:

Collected and curated human protein data with known O-linked threonine glycosylation sites; encoded sequences with seven feature coding methods; trained a random forest classifier after algorithm evaluation; validated using 5-fold cross-validation and an independent validation dataset.