GPCR-PTM: an interpretable, topology-aware rule-based framework for residue-level post-translational modification annotation and prediction across the human GPCRome

@article{chen2026gpcrptm,
  title = {{GPCR-PTM: an interpretable, topology-aware rule-based framework for residue-level post-translational modification annotation and prediction across the human GPCRome}},
  author = {Yinzhen Chen and Shiyu Wang and Shuguang Yuan},
  journal = {Medicine in Drug Discovery},
  year = {2026},
  month = oct,
  doi = {10.1016/j.medidd.2026.100267},
  url = {https://doi.org/10.1016/j.medidd.2026.100267},
  note = {CC BY},
}

TY  - JOUR
TI  - GPCR-PTM: an interpretable, topology-aware rule-based framework for residue-level post-translational modification annotation and prediction across the human GPCRome
AU  - Yinzhen Chen
AU  - Shiyu Wang
AU  - Shuguang Yuan
PY  - 2026
DA  - 2026/10/30
JO  - Medicine in Drug Discovery
DO  - 10.1016/j.medidd.2026.100267
UR  - https://doi.org/10.1016/j.medidd.2026.100267
AB  - Post-translational modifications (PTMs) are major regulators of G protein-coupled receptor (GPCR) signaling, trafficking and surface expression, yet residue-level evidence is fragmented across databases and computational predictions often ignore the compartment constraints imposed by the seven-transmembrane topology. We developed GPCR-PTM, an interpretable framework that integrates database-backed annotations with topology-constrained, rule-based prediction of phosphorylation, N-linked glycosylation, S-palmitoylation and ubiquitination. Sites are separated into Verified (UniProt ECO:0000269 experimental annotation), Supported (other database-backed evidence) and Predicted (rule-based) evidence layers; database citations are screened at the PubMed title/abstract level; and predicted sites are ranked using explicit sequence rules, membrane topology and evolutionary conservation. Across 839 reviewed human GPCRs, the pipeline consolidated 9559 source records into 7247 unique database-backed PTM sites spanning 827 receptors. Among 13,453 merged citation-site pairs, 13,216 (98.2%) had retrievable cached abstracts, but only 353 (2.7%) were residue-explicit, showing that abstract granularity should not be equated with evidence quality. In a retrospective benchmark against 246 UniProt ECO:0000269 experimental sites, the rule system recovered 216 sites (87.8%), including 75.4% of phosphorylation sites and all evaluable N-linked glycosylation and S-palmitoylation sites. Because the glycosylation and palmitoylation rules encode the canonical recognition patterns themselves, complete recovery for these classes reflects rule coverage rather than predictive discrimination; phosphorylation, the least pattern-bound class, reached a strict precision of 7.0% and an AUPRC of 0.074, so the overall recovery should be read as coverage under blinded lookup rather than calibrated predictive performance. Compartment-matched enrichment further supported the pXpp phosphorylation cluster (3.19-fold enrichment) and confirmed the expected localization of canonical N-glycosylation and juxtamembrane palmitoylation rules. Interactive bilingual reports combine evidence filtering, a PTM landscape overview, topology-aware sequence annotation and AlphaFold-based 3D context. GPCR-PTM is intended as an auditable hypothesis-generation resource for basic GPCR research and pharmacology rather than a calibrated probability predictor. Prospective validation and direct comparison with learned predictors will be required to quantify predictive performance on previously unseen biology.
SN  - 2590-0986
ER  -