Use this link to cite:
https://hdl.handle.net/2183/49627 Priorización de variantes genómicas asociadas a Fibrilación Ventricular mediante aprendizaje automático
Loading...
Identifiers
Publication date
Authors
Fole Abellás, Hugo
Other responsabilities
Universidade da Coruña. Facultade de Informática
Journal Title
Bibliographic citation
Type of academic work
Academic degree
Abstract
[Resumen] La Fibrilación Ventricular idiopática es la principal causa de muerte súbita cardíaca en adultos y su base genética permanece en gran medida desconocida. El objetivo de este trabajo es identificar variantes genómicas candidatas en una cohorte reducida, proporcionando a los clínicos un conjunto razonado de posiciones sobre las que centrar sus estudios funcionales.
Para ello se desarrolla un pipeline que combina selección de características con Fast-mRMR en validación cruzada anidada y un clasificador XGBoost con interpretabilidad mediante valores SHAP out-of-fold. El resultado principal es una lista priorizada de variantes con plausibilidad biológica, obtenida reduciendo el espacio de búsqueda de más de 120.000 posiciones a un conjunto de consenso de menos de 500.
[Abstract] Idiopathic Ventricular Fibrillation is the leading cause of sudden cardiac death in adults and its genetic basis remains largely unknown. The goal of this work is to identify candidate genomic variants in a small cohort, providing clinicians with a reasoned set of positions on which to focus their functional studies. A pipeline is developed combining Fast mRMR feature selection within nested cross-validation and an XGBoost classifier with out-of-fold SHAP interpretability. The main result is a prioritised variant list with potential biological plausibility, obtained by reducing the search space from over 120,000 positions to a consensus set of fewer than 500.
[Abstract] Idiopathic Ventricular Fibrillation is the leading cause of sudden cardiac death in adults and its genetic basis remains largely unknown. The goal of this work is to identify candidate genomic variants in a small cohort, providing clinicians with a reasoned set of positions on which to focus their functional studies. A pipeline is developed combining Fast mRMR feature selection within nested cross-validation and an XGBoost classifier with out-of-fold SHAP interpretability. The main result is a prioritised variant list with potential biological plausibility, obtained by reducing the search space from over 120,000 positions to a consensus set of fewer than 500.
Description
Editor version
Rights
Attribution 4.0 International







