BIOMARKER SELECTION FOR PEDIATRIC SEPSIS DIAGNOSIS USING DEEP LEARNING

Abstract

This study proposes a new approach

for diagnosing pediatric sepsis that utilizes a convolutional

neural network and a combination of 7 immune-related

genes (IRGs), including CD24, TTK, PRG2, CLEC7A,

CCL3, TNFAIP3, and CCRL2. A three-layer gene selection

process involves sequential procedure that combines

differential gene expression analysis of immune-related

genes with a gene score calculation using the F-score

algorithm to identify the most informative differentially

expressed genes, and then using deep learning models to

identify the optimal gene combination. The performance

of the proposed algorithm is evaluated using a 3-fold

cross-validation procedure with deep learning models. The

results show that the selected gene combinations achieve an

accuracy of 91.92% and an area under the ROC curve of

87.86%, indicating that the proposed algorithm is reliable

for predicting pediatric sepsis mortality. Additionally, the

identification of a signature consisting of 7 IRGs associated

with pediatric sepsis mortality has the potential to aid in

the development of dependable diagnostic and prognostic

biomarkers for sepsis