Construction of a diagnostic model for temporal lobe epilepsy using interpretable deep learning: disease-associated...

Area of Science:

• Neurology
• Computational Biology
• Genetics

Background:

• Temporal lobe epilepsy (TLE) is a complex neurological disorder with poorly understood genetic factors.
• Accurate diagnosis and identification of disease markers are crucial for effective TLE management.

Purpose of the Study:

• To develop an interpretable deep learning model for TLE diagnosis.
• To identify key genes associated with TLE pathogenesis using advanced computational methods.

Main Methods:

• RNA-sequencing and microarray data from 287 samples across eight GEO datasets were analyzed.
• Multiple machine learning algorithms, including Deep Neural Networks (DNN), were trained to distinguish TLE from normal samples.
• SHapley Additive exPlanations (SHAP) and Kolmogorov-Arnold Networks (KAN) were used for model interpretability and gene identification.

Main Results:

• A DNN model achieved perfect diagnostic performance (AUC = 1.000, accuracy = 1.000) using 10 optimized genetic features.
• SHAP analysis identified DEPDC5, STXBP1, GABRG2, SLC2A1, and LGI1 as significant TLE-associated genes.
• An online diagnostic platform was developed for clinical application.

Conclusions:

• The study presents a transparent and interpretable deep learning model for TLE diagnosis.
• This model and identified genes offer a supplementary tool to improve TLE diagnostic accuracy in clinical settings.