No abstract

Background Due to multiple pathogenesis of Alzheimer’s disease (AD), currently discovered biomarkers are stilled limited for its classification and diagnosis, robust and universal biomarkers or biomarker combinations need further to be explored. Methods Based on machine learning, The SVM-RFECV algorithm screened out a 12-protein panel that was applied to 5 different cohorts of AD cerebrospinal fluid (CSF) proteomic datasets. Results The 12-protein panel exhibited strong diagnosibility and high accuracy. It was involved in several AD related biological process and highly correlated with classical AD pathogenic biomarkers (Aβ, tau/p-tau and Montreal Cognitive Assessment (MoCA) score). It was also capable of distinguishing early stage of AD (mild cognitive impairment, MCI) as well as from other neurodegenerative diseases. Conclusions The SVM-RFECV algorithm has great advantages of robust predicting ability, high accuracy and good reliability for identifying AD, thus providing clues for AD pathogenesis and shedding light on AD diagnosis clinically.

No abstract