A MACHINE LEARNING FRAMEWORK for the DETECTION of ALZHEIMER’S DISEASE USING GUT FUNGAL ITS SEQUENCING DATA

Abstract

Alzheimer's disease is a common neurodegenerative disease. The microbiota-gut brain axis has emerged as a paradigm in the understanding of AD, and attention has been focused mainly on the bacterial microbiome, whereas the immunologically potent gut mycobiome (fungal community) has been largely overlooked. Most research on the mycobiome associated with AD has been descriptive with little predictive clinical application. To overcome this methodological limitation, this study aimed to create a supervised machine learning model to distinguish between AD patients and cognitively normal controls based on non-invasive gut fungal Internal Transcribed Spacer (ITS) sequencing data. A public dataset of 120 individuals (60 AD patients and 60 healthy controls) was used for the research. Raw ITS amplicon sequencing data was processed in a standardized, bioinformatics pipeline using QIIME 2 and the DADA2 algorithm for high resolution Amplicon Sequence Variant (ASV) generation and the UNITE database for taxonomic assignment. Four machine learning models: Logistic Regression, Support Vector Machine, Histogram-based Gradient Boosting and Random Forest were optimized and evaluated using a rarefied, high-dimensional ASV feature matrix. The results showed that the Random Forest classifier had the highest classification accuracy (0.75) and ROC-AUC (0.85) on an independent test set. Linear models on the other hand, did not work very well, showing the complexity and the non-linearity of the microbial interactions. The Random Forest model was shown to be stable by rigorous 5-fold cross validation. In addition, the feature importance analysis showed that certain genera of fungi such as Candida, Purpureocillium, and Oligophagozyma were the most discriminative predictive biomarkers in AD. The strong predictive value of Candida is in line with the immunological mechanisms of the gut-brain axis and inflammatory pathogenesis. Finally, this study moves the mycobiome analysis from an ecological description to a predictive clinical informatic, making fungal profiles of the gut a new potential non-invasive diagnostic tool for Alzheimer's disease.