A Machine Learning Classifier for Predicting Stable MCI Patients Using Gene Biomarkers

Alzheimer's disease (AD) is a neurodegenerative disorder with an insidious onset and irreversible condition. Patients with mild cognitive impairment (MCI) are at high risk of converting to AD. Early diagnosis of unstable MCI patients is therefore vital for slowing the progression to AD. However, current diagnostic methods are either highly invasive or expensive, preventing their wide applications. Developing low-invasive and cost-efficient screening methods is desirable as the first-tier approach for identifying unstable MCI patients or excluding stable MCI patients. This study developed feature selection and machine learning algorithms to identify blood-sample gene biomarkers for predicting stable MCI patients. Two datasets obtained from the Alzheimer's Disease Neuroimaging Initiative (ADNI) database were utilized to conclude 29 genes biomarkers (31 probes) for predicting stable MCI patients. A random forest-based classifier performed well with area under the receiver operating characteristic curve (AUC) values of 0.841 and 0.775 for cross-validation and test datasets, respectively. For patients with a prediction score greater than 0.9, an excellent concordance of 97% was obtained, showing the usefulness of the proposed method for identifying stable MCI patients. In the context of precision medicine, the proposed prediction model is expected to be useful for identifying stable MCI patients and providing medical doctors and patients with new first-tier diagnosis options.

Automated summary

This study developed blood-sample gene biomarkers to predict stable MCI patients using feature selection and machine learning algorithms. Utilizing datasets from the Alzheimer's Disease Neuroimaging Initiative (ADNI), 29 gene biomarkers (31 probes) were identified for predicting stable MCI patients. A random forest-based classifier showed good performance with AUC values of 0.841 and 0.775 for cross-validation and test datasets, respectively, and achieved 97% concordance for patients with prediction score > 0.9.