Diagnosis of Epilepsy by Machine Learning of High-Performance Plasma Metabolic Fingerprinting

Epilepsy is a chronic neurological disorder that causes a major threat to public health and the burden of disease worldwide. High-performance diagnostic tools for epilepsy need to be developed to improve diagnostic accuracy and efficiency while still missing. Herein, we utilized nanoparticle-enhanced laser desorption/ionization mass spectrometry (NELDI MS) to acquire plasma metabolic fingerprints (PMFs) from epileptic and healthy individuals for timely and accurate screening of epilepsy. The NELDI MS enabled high detection speed (∼30 s per sample), high throughput (up to 384 samples per run), and favorable reproducibility (coefficients of variation < 15%), acquiring high-performed PMFs. We next constructed an epilepsy diagnostic model by machine learning of PMFs, achieving desirable diagnostic capability with the area under the curve (AUC) value of 0.941 for the validation set. Furthermore, four metabolites were identified as a diagnostic biomarker panel for epilepsy, with an AUC value of 0.812-0.860. Our approach provides a high-performed and high-throughput platform for epileptic diagnostics, promoting the development of metabolic diagnostic tools in precision medicine.