Stratifying Mild Cognitive Impairment Patients via Embedded Transcriptomic Data and Clustering Analysis

Alzheimer's disease (AD) is a complex brain disorder that causes progressive cognitive decline and neuropsychiatric complications, ranking among the top ten causes of death. Clinically, there are three stages of cognitive decline: dementia, mild cognitive impairment (MCI), and subjective cognitive decline. Both MCI and subjective cognitive decline are recognized as risk factors for dementia, but most countries do not recommend specific treatments outside of clinical trials. A key goal of precision medicine is to improve the classification of pathological conditions based on molecular markers rather than relying solely on clinical symptoms. From this perspective, providing a clear and objective identification of conditions that could progress into more severe symptoms while distinguishing them from other types of impairments is crucial for improving early detection of AD. In this work, we propose a clustering-based molecular stratification of MCI patients by analyzing blood transcriptomic data. Unsupervised and supervised approaches were adopted to select features from gene expression data, and deep learningbased autoencoders were applied to embed the data into a lower-dimensional latent space. This latent representation was then fed into a clustering algorithm to determine the final result. The identified MCI patient clusters showed significant molecular, clinical, and pathophysiological differences. Differences were also detected by comparing MCI patient clusters to AD patients and control samples. These findings can lead to an early and precise diagnosis of progressive dementia, offering a more objective alternative to standard cognitive-based definitions.