# Prediction of mild cognitive impairment using blood multi-omics data

## TL;DR

This study uses blood-based genomic data to predict mild cognitive impairment with high accuracy, offering a non-invasive diagnostic tool.

## Contribution

The study is the first to show that genome structure data like CNVs can be as informative as gene expression for MCI prediction.

## Key findings

- An XGBoost model achieved an AUC of 0.9398 using gene expression and CNV data from blood samples.
- 149 genomic features important for MCI prediction were identified and linked to neurodegenerative pathways.
- Blood-based multi-omics data effectively distinguish MCI patients from normal controls.

## Abstract

Mild cognitive impairment (MCI) represents an initial phase of memory or other cognitive function decline and is viewed as an intermediary stage between normal aging and Alzheimer’s disease (AD), the most prevalent type of dementia. Individuals with MCI face a heightened risk of progressing to AD, and early detection of MCI can facilitate the prevention of such progression through timely interventions. Nonetheless, diagnosing MCI is challenging because its symptoms can be subtle and are easily missed. Using genomic data from blood samples has been proposed as a non-invasive and cost-efficient approach to build machine learning predictive models for assisting MCI diagnosis. However, these models often exhibit poor performance. In this study, we developed an XGBoost-based machine learning model with AUC (the Area Under the receiver operating characteristic Curve) of 0.9398 utilizing gene expression and copy number variation (CNV) data from patient blood samples. We demonstrated, for the first time, that data at a genome structure level such as CNVs could be as informative as gene expression data to classify MCI patients from normal controls. We identified 149 genomic features that are important for MCI prediction. Notably, these features are enriched in the pathways associated with neurodegenerative diseases, such as neuron development and G protein-coupled receptor activity. Overall, our study not only demonstrates the effectiveness of utilizing blood sample-based multi-omics for predicting MCI, but also provides insights into crucial molecular characteristics of MCI.