PhD Scientific Days 2026

Poster Session 3.W - Pharmaceutical Sciences and Health Technologies

Identifying Biomarkers Associated with Long COVID Subtypes Using Artificial Intelligence Models

Előadó neve

Bereczki, Zoltán, MSc

Neptune code

E9BX9T

Előadó munkahelye

Department of Pharmacology and Pharmacotherapy, Semmelweis University

Előadó telefonszáma

06307128401

Előadó e-mail címe

bereczki.zoltan.andras@semmelweis.hu

Az előadás címe

Identifying Biomarkers Associated with Long COVID Subtypes Using Artificial Intelligence Models

Szerző(k) neve és munkahelye

Zoltán Bereczki^1,2, Olivér Márton Balogh^1,2, Dominika Lukovic³, Antonia Domanig³, Roland Molontay⁴, Péter Ferdinandy^1,2,5, Bence Ágg^1,2,5, Mariann Gyöngyösi³

1: Department of Pharmacology and Pharmacotherapy, Semmelweis University, H-1085 Budapest, Hungary
2: Center for Pharmacology and Drug Research & Development, Semmelweis University, H-1085 Budapest, Hungary
3: Division of Cardiology, Department of Internal Medicine II, Medical University of Vienna, 1090 Vienna, Austria
4: Institute of Biostatistics and Network Science, Semmelweis University, Üllői út 26., Budapest, H-1085, Hungary
5: Pharmahungary Group, H-6722 Szeged, Hungary

Bemutatás módja

Poszter

Szekció

Poster Session 3.W - Pharmaceutical Sciences and Health Technologies

Language of the presentation

English

Preferred session

Pharmaceutical Sciences and Health Technologies

Összefoglaló szövege

Introduction
Long COVID multiorgan disease, a long-term consequence of COVID-19 infection with a prevalence of about 3-20% represents a major challenge to healthcare systems worldwide. In the absence of diagnostic biomarkers and specific therapies, early diagnosis is crucial to prevent disease progression. Accordingly, biomarkers and computational approaches for distinguishing between the clinical phenotypes of long COVID patients (cardiovascular, pulmonary or neurological) are still lacking.
Aims
Our aim was to identify potential biomarkers associated with long COVID subtypes through developing an artificial intelligence (AI) model trained to distinguish between these subtypes, using clinical data.
Methods
The dataset of 544 long COVID patients included 864 features. After preprocessing, features were selected by various missingness thresholds, and missing values in the remaining features were imputed using several methods. Decision-tree-based AI models (XGBoost, RandomForest, and CART) were trained on the resulting imputed datasets and evaluated using k-fold cross-validation with accuracy, weighted F1, and AUROC. Feature importance was assessed using SHapley Additive exPlanations (SHAP).
Result
After data preprocessing, 517 long COVID patients and 725 features were kept for further analysis. According to preliminary results, RandomForest model performed best (mean AUROC of 0.77 ± 0.03) with a maximum of 40% feature missingness. Feature importance analysis revealed potential biomarkers. For example, non-physiological cardiac abnormalities detected by either echocardiography or MRI, pulmonary symptoms, and absence of cardiovascular complaints are strongly associated with increased probability of cardiovascular, pulmonary, neurological subtype classification, respectively.
Conclusion
This is the first demonstration of identifying biomarker candidates characteristic of cardiovascular, pulmonary, and neurological subtypes of long COVID as a multi-organ disease, using a RandomForest-based AI model trained on clinical data from long COVID patients.
Funding
Project no. RRF-2.3.1-21-2022-00003 has been implemented with the support provided by the European Union. B.Z has been supported by a Semmelweis 250+ Excellence Fellowship. Gy.M has been supported by the Austrian Science Fund KLI 1064-B and Medical Scientific Fund of the Mayor of the City of Vienna 21176.

University

Semmelweis University

Supervisor

Bence Ágg

Publication of my abstract

I do not give consent to the publication of my abstract on the website of the congress.

phd.section.field

in doctoral studies before complex exam (PhD)

Kind

Szabad

Status

elfogadva

Accepted presentation method

poszter

Előadás fájl jóváhagyás

nem rendelkezett róla

Előadó

9757

Start

13:36

End

13:39