Deployed model documentation

O-GlcNAcPRED-DL model card

Scope, outputs, limitations, and reproducibility information for version 1.0.0.

Model and intended use

The deployed O-GlcNAcPRED-DL 1.0.0 release is the published species-specific ensemble for prioritizing candidate serine and threonine O-GlcNAcylation sites in human or mouse protein sequences. It supports experimental planning; it does not establish modification.

Inputs and outputs

Input is one or more protein sequences in FASTA format with a selected human or mouse model. The browser reports each S/T residue, its one-based position, a model score rounded to three decimals, and the published confidence thresholds: greater than 0.5, 0.95, and 0.99.

Architecture and runtime

Five CNN–BiLSTM models combine AAindex and Word2Vec sequence features with published species-specific ensemble weights. TensorFlow.js/WASM inference runs locally in a Web Worker. The tracked manifest identifies every model and feature artifact by SHA-256.

Limitations

  • Performance evidence is strongest for human and mouse, the two trained species.
  • A high score is not experimental evidence, and a low score does not prove absence.
  • The score should not be interpreted as a universally calibrated biological probability.
  • Sequence context alone does not represent tissue, condition, localization, enzyme availability, or measurement bias.
  • Unusual residues and very large requests are restricted by the documented browser input contract.

Validation and reproducibility

The browser implementation is tested against Python-generated golden results for both species. The published study provides the model-development and comparative evaluation context. View the complete citation.

A prospective PRED-DL 2.0 protocol exists in the source repository, but no v2 model is claimed or deployed before its future corpus freeze and release gates pass.