Related Experiment Video
Updated: Sep 9, 2025
08:51
Statistical Modelling of Cortical Connectivity Using Non-invasive Electroencephalograms
Published on: November 1, 2019
5.7K
Interpretable EEG biomarkers for neurological disease models in mice using bag-of-waves classifiers
Maria Isabel Cano Achuri1, Montana Kay Lara2,3, Khalil Abed Rabbo4
1Department of Electrical and Computer Engineering, University of Delaware, Newark, Delaware, USA.
Biorxiv : the Preprint Server for Biology
|September 2, 2025
Summary
Machine learning accurately predicts epilepsy genotypes from mouse brain activity (Electroencephalograms/EEGs). This method identifies neurological disease-genotypes in mice without overt seizures, using EEG waveform patterns as biomarkers.
Area of Science:
- Neuroscience
- Genetics
- Computational Biology
Background:
- Electroencephalograms (EEGs) record brain's electrical activity, showing patterns linked to specific genotypes.
- Identifying neurological disease-genotypes, especially for epilepsy, is challenging without direct seizure observation.
- Subtle neurological phenotypes in animal models often lack overt epilepsy, complicating genotype diagnosis.
Purpose of the Study:
- To investigate the prediction of neurological disease-genotypes from long-term EEG signals in mice.
- To develop a machine learning approach for extracting EEG waveform biomarkers for genotype prediction.
- To assess the efficacy of using EEG waveform occurrence counts for identifying epilepsy-associated genotypes.
Main Methods:
- Utilized long-term EEG signals from freely behaving mice across six groups (three inbred strains with and without TSC1 gene knockout).
- Developed a machine learning model using a dictionary of optimized waveforms to approximate EEG windows.
- Employed logistic regression on waveform occurrence counts as features for genotype prediction, validated through cross-validation.
Main Results:
- Waveform counts from multi-hour EEG segments reliably predicted mouse strain (70% accuracy).
- Strain-specific classifiers identified the epilepsy-genotype (TSC1 knockout) with 67% sensitivity in DBA2 and C57B6 mice.
- High specificity was achieved (100% for DBA2, 67% for C57B6) without overt seizures or detected seizures.
Conclusions:
- EEG waveforms serve as valuable phenotypical biomarkers for identifying epilepsy genotypes.
- The 'bag-of-waves' feature representation is effective for genotype prediction from EEG data.
- This machine learning approach holds potential for diagnosing subtle neurological phenotypes and epilepsy genotypes.
