# Plasma Proteomic Signatures for Alzheimer's Disease: Comparable Accuracy to ATN Biomarkers and Cross‐Platform Validation

## TL;DR

This study identifies a plasma protein signature for Alzheimer's disease that performs as well as traditional biomarkers and works across different testing platforms.

## Contribution

A cross-platform validated plasma proteomic signature for Alzheimer's disease with high accuracy and fewer analytes.

## Key findings

- An 11-analyte signature achieved 93.5% accuracy in distinguishing Alzheimer's from cognitively normal individuals.
- The signature maintained 95.2% accuracy when validated on a different proteomic platform and population.
- The MCI classification model had lower accuracy and failed to reliably distinguish decliners from non-decliners.

## Abstract

There is growing recognition of the potential of plasma proteomics for Alzheimer's Disease (AD) risk assessment and disease characterization. However, differences between proteomics platforms introduce uncertainties regarding cross‐platform applicability.

We aimed to identify a detailed plasma biosignature for distinguishing AD from cognitively normal (CN) and another signature for classifying mild cognitive impairment (MCI) decliners and non‐decliners. We also explored the cross‐platform applicability of these models between two proteomic platforms.

Elastic net was performed on 190 plasma analytes measured using the Luminex xMAP platform in 566 participants from the Alzheimer's Disease Neuroimaging Initiative (ADNI) to model MCI stable/decliner and AD/CN classification. MCI decliner was defined as progression to AD during follow‐up (mean 4.2 ± 3.2 years). External cross‐platform validation was conducted with 1303 participants from the Spanish Ace study, using the SOMAscan 7k platform.

An 11‐analyte signature for distinguishing AD from CN achieved a 93.5% accuracy on ADNI and 95.2% on Ace. The ApoE and BNP proteins were the two most important contributors to the classifier. The MCI classification signature performed less well, with 65.9% accuracy on ADNI and 51.0% accuracy upon validation testing in Ace.

Compared with prior proteomic‐based studies on the same dataset, our findings attained higher specificity and sensitivity for AD classification while utilizing a smaller panel of analytes. We also confirmed the reliability and consistency of this signature within a different population from a different platform. The plasma proteomic platforms explored were, however, not sufficient to determine MCI decliners versus non‐decliners.