# Consensus machine learning identifies cell death gene signature for carotid artery stenosis diagnosis

## TL;DR

A machine learning approach identified a three-gene signature that could improve early diagnosis of carotid artery stenosis.

## Contribution

A novel three-gene diagnostic signature for CAS was developed using consensus machine learning and multi-omics validation.

## Key findings

- Integration of nine datasets identified 14 cell death-related genes consistently associated with CAS.
- The MLDS (IRF1, FYCO1, FDFT1) showed high cross-cohort diagnostic performance.
- FYCO1 downregulation was confirmed in plaques and blood, with impaired autophagy and increased inflammation.

## Abstract

Carotid artery stenosis (CAS) is a major contributor to ischemic stroke, and molecular tools for its early detection remain limited. To address this need, we integrated one in-house RNA-seq cohort with eight public datasets comprising 696 samples, together with proteomic profiling, RT-qPCR, single-cell sequencing, and FYCO1 silencing experiments. From 1,258 curated cell death-related genes, candidates were filtered by logistic regression across cohorts, and ten machine learning algorithms were combined into 105 model configurations to derive a consensus diagnostic classifier. Fourteen genes showed consistent associations with CAS, and the machine learning-derived diagnostic signature (MLDS), consisting of IRF1, FYCO1, and FDFT1, demonstrated the highest cross-cohort performance. FYCO1 downregulation was validated in plaques and blood and supported by single-cell analysis, while functional assays indicated impaired autophagic flux and heightened inflammatory signaling. These findings highlight MLDS as a robust molecular tool that may enhance the precision diagnosis of CAS.

•Nine cohorts (696 samples) systematically integrated for CAS biomarker discovery•105 ML model configurations yielded a consensus three-gene diagnostic signature•Transcriptomic, proteomic, single-cell, and functional assays validated FYCO1•MLDS offers a non-invasive tool for early CAS detection and precision medicine

Nine cohorts (696 samples) systematically integrated for CAS biomarker discovery

105 ML model configurations yielded a consensus three-gene diagnostic signature

Transcriptomic, proteomic, single-cell, and functional assays validated FYCO1

MLDS offers a non-invasive tool for early CAS detection and precision medicine

Health sciences