Abstracts

Machine Learning EEG Biomarkers in SYNGAP1 Rodent Models and Patients

Abstract number : 1.213
Submission category : 2. Translational Research / 2C. Biomarkers
Year : 2024
Submission ID : 1265
Source : www.aesnet.org
Presentation date : 12/7/2024 12:00:00 AM
Published date :

Authors :
Melissa Fasol, PhD – University of Edinburgh
Javier Escudero, PhD – University of Edinburgh
Presenting Author: Alfredo Gonzalez-Sulser, PhD – University of Edinburgh


Rationale: Mutations in SYNGAP1, a gene responsible for regulating synaptic function, are reported to account for up to 1% of neurodevelopmental disorders. Most patients with pathogenic SYNGAP1 variants experience absence seizures, which coincide with developmental delay. The identification of non-invasive biomarkers which assess synaptic function could be invaluable to diagnose patients, track disease progression and determine treatment efficacy. We recently reported that a rat model of SYNGAP1 haploinsufficiency displays spontaneous seizures, abnormal social interactions, lack of extinction of fear learning, irregular sleep dynamics, and reduced connectivity between EEG electrodes.


Methods: We analysed EEG recordings from SYNGAP1 heterozygous mutant rats and littermate controls, as well as overnight recordings from human patients with SYNGAP1 mutations and sibling controls. For individual EEG recording epochs excluding seizures (5 seconds in rats, 30 seconds in humans), we calculated feature values for signal complexity, spectral analysis and functional connectivity. We trained an extreme gradient boosting (XGBoost) machine learning classifier to differentiate between SYNGAP1 mutants and controls. We then applied the SHapley Additive exPlanations (SHAP) analysis to the classifier’s predictions to determine which features were critical for identification


Results: Using the XGBoost classifier we obtained out-of-the-box accuracy, precision, recall and F1 scores of 75%, 76%, 76% and 71% in rats and, 82%, 98%, 72% and 83% in humans respectively. The SHAP analysis revealed that functional connectivity metrics derived from somatosensory regions played a prominent role in detecting SYNGAP1 haploinsufficiency.


Conclusions: EEG machine learning analysis can efficiently segregate data from mutants and controls efficiently in both rats and humans. Connectivity parameters are most critical for machine identification in both species, suggesting animal model validity and potential clinical applicability.


Funding: Simons Initiative for the Developing Brain, Oracle for Research, and The Carnegie University Trust

Translational Research