eng Oriental Scientific Publishing Company Biosciences Biotechnology Research Asia 0973-1245 2026-09-14 23 3 59706 article Leeshma Koroth 1 Praveena Marannan 1 Department of Computer Science, Dr. S. N. S. Rajalakshmi College of Arts and Science, Coimbatore, India Epilepsy is a neurological disorder with varied symptoms and clinical heterogeneity, and its early diagnosis remains difficult in clinical practice. Current machine learning and deep learning methods mainly focus on handcrafted EEG features and traditional sequential models. They are not able to effectively utilize unstructured Electronic Health Records (EHRs) to learn long-range semantic dependencies and latent biomedical relationships. In this paper, An Explainable Biomedical Graph Transformer Model (BioClinX-GraphFormer) is presented to automatically predict epilepsy phenotypes from EHRs. Here, the prediction task is formulated as a binary phenotype classification (Epilepsy vs. Non-Epilepsy and Seizure vs. Non-Seizure). The proposed method combines NLP, ClinicalBERT contextual embeddings, Biomedical Knowledge Graph construction, and Graph Transformer Networks to simultaneously learn semantic and relational clinical dependencies. The clinical notes acquired from MIMIC-IV and seizure-associated EEG reports from CHB-MIT are pre-processed using tokenization, lemmatization, NER, and clinical entity normalization linking. Seizure symptoms, drugs, EEG descriptors, and neurological terms extracted from clinical notes are converted to graphs and passed through multi-head attention graphs to predict the epilepsy phenotype. The proposed BioClinX-GraphFormer is benchmarked against SVM, RF, XG-Boost, NLP-BiLSTM, and Contextual Transformer-Augmented ClinicalBERT models. Results show promising performance with an accuracy of 98.1%, F1-score of 98.0% and an AUC-ROC of 0.985. The proposed work combines explainable biomedical graphs with contextual semantic learning to provide scalable, interpretable, and trustworthy binary epilepsy phenotype predictions. https://www.biotech-asia.org/vol23no3/explainable-biomedical-graph-transformer-for-automated-epilepsy-phenotype-prediction-from-ehr-data/ Biomedical Knowledge Graph; ClinicalBERT; Electronic Health Records; Explainable Artificial Intelligence; Graph Transformer Networks; Seizure Prediction and Classification