JMIR Publications

ABSTRACT

Background:

Drug-related seizures require timely recognition in complex treatment settings, but spontaneous-report associations, whole-molecule features, and network-structural clues are rarely examined together.

Objective:

To characterize drug-related seizures by integrating pharmacovigilance, high-dimensional chemical space, machine-learning applicability, and independent network and structural analyses.

Methods:

We curated FAERS data from 2004Q1 through 2026Q1 and defined seizures using SEIZURE plus historical CONVULSION. After deduplication and ingredient mapping, ROR, PRR, BCPNN, and MGPS were evaluated across drug-role and indication-exclusion specifications. We analyzed ring topology, Morgan fingerprints, and USRCAT space, then audited a benchmark XGBoost model using an applicability domain and refits. Separately, 12 neuro-oncology drugs entered network analysis; FAERS-positive drugs related to at least 2 hubs underwent docking and one 100-ns molecular-dynamics trajectory per complex.

Results:

The cohort contained 167,223 unique seizure-related reports. Auditing retained 68 of 844 eligible ingredients across specifications. Morgan and USRCAT spaces showed same-label enrichment in all 8 tests (empirical P=1/10,001; all Benjamini-Hochberg significant). XGBoost held-out accuracy was 0.783 and Active recall was 0.364. Eleven of 12 primary drug structures were in-domain, although 7 had unseen Murcko scaffolds. The 486 drug-associated targets and 3951 epilepsy-associated genes shared 236 genes, yielding 15 exploratory hubs. Independently selected everolimus and vincristine were screened against human HSP90AA1 and the 5EWM cross-species interface comprising Xenopus laevis GRIN1 chain C and human GRIN2B chain D. Four trajectories showed differential behavior; vincristine-HSP90AA1 had the most consistently low RMSD profile.

Conclusions:

Integrating reporting robustness, whole-molecule organization, model-scope auditing, and independent network-structural analysis produced a traceable strategy for prioritizing drug-context combinations and testable molecular hypotheses.