Machine learning-driven serum metabolomics: insights into the transition to early-stage heart failure with preserved ejection fraction
Circulation
; 146(Suppl 1)Nov 8, 2022. ilus
Artigo
em Inglês
| CONASS, Sec. Est. Saúde SP, SESSP-IDPCPROD, Sec. Est. Saúde SP
| ID: biblio-1399709
Biblioteca responsável:
BR79.1
ABSTRACT
Introduction:
Metabolomics has emerged as a powerful tool in providing readouts of early disease states before clinical manifestation. Here we used the predictive power of Unsupervised Hierarchical Clustering Analysis (UHCA) and Automated Machine Learning (AutoML) algorithms to identify serum metabolic panels in a population at risk of developing HFpEF.Methods:
We studied 215 subjects staged as non-HF, pre-HFpEF and early-stage HFpEF(es-HFpEF). We evaluated clinical, laboratory, echocardiographic, and NMR-based metabolomics of blood serum data. UHCA and AutoML were used to explore metabolic fingerprints potentially related to clinical features or HFpEF. We used Metabolite Set Enrichment Analysis to explore biochemical pathways.Results:
The UHCA identified three major patients (P) and two metabolites (M) clusters (Figure). The P clusters were associated with HFpEF stages, cardiac remodeling, diastolic dysfunction, and sex (Pearson Chi-square, p < 0.05) and M clusters with glycine and serine metabolism and urea cycle pathways (FDR-adjusted p-value < 0.002). Considering non-HFpEF and es-HFpEF groups, AUROC mean for feature subset combinations was 0.897 and the highest AUROC (0.995) combined metabolites, clinical, laboratory and echo features. Of the 64 models trained that included metabolites as input, serine (25), uridine (17), 2-oxoglutarate (14), citrate (14), 2-aminobutyrate (13) and taurine (13) were observed more frequently with feature importance value greater than zero. The metabolites with higher sum values of feature importance were serine (0.173), uridine (0.131), 2-aminobutyrate (0.123), choline (0.098) and dimethylamine (0.087).Conclusions:
This study revealed characteristic metabolite profiles in the sera of patients at risk of developing HFpEF. These metabolite panels can add information for classificatory algorithms development and contribute to the understanding of HFpEF pathophysiology.
Texto completo:
Disponível
Coleções:
Bases de dados nacionais
/
Brasil
Base de dados:
CONASS
/
Sec. Est. Saúde SP
/
SESSP-IDPCPROD
Assunto principal:
Fatores de Risco
/
Insuficiência Cardíaca Diastólica
/
Aprendizado de Máquina
/
Insuficiência Cardíaca
Tipo de estudo:
Estudo de etiologia
/
Estudo prognóstico
/
Fatores de risco
Idioma:
Inglês
Revista:
Circulation
Ano de publicação:
2022
Tipo de documento:
Artigo
/
Congresso e conferência
Instituição/País de afiliação:
Instituto Dante Pazzanese de Cardiologia/BR