Machine learning-driven identification of the gene-expression signature associated with a persistent multiple organ...

Area of Science:

• Genomics
• Critical Care Medicine
• Bioinformatics

Background:

• Multiple Organ Dysfunction Syndrome (MODS) is a major cause of death in critically ill patients, but its underlying biological mechanisms remain poorly understood.
• Identifying genes associated with persistent MODS trajectories is crucial for understanding its pathobiology and predicting patient risk.

Purpose of the Study:

• To identify a set of genes associated with a persistent MODS trajectory using supervised machine learning.
• To develop and validate a predictive model for persistent MODS in critically ill patients.

Main Methods:

• Secondary analysis of publicly available gene-expression datasets.
• Application of supervised machine learning (ML) to identify genes in pediatric septic shock patients.
• Optimization and validation of the ML model in independent datasets.

Main Results:

• A persistent MODS trajectory was characterized by 568 differentially expressed genes and a dysregulated innate immune response.
• Supervised ML identified 111 genes associated with persistent MODS, achieving an AUROC of 0.87 in the training set.
• An optimized 20-gene model predicted persistent MODS with AUROCs of 0.74-0.79 in validation and test sets, irrespective of host age or cause of organ dysfunction.

Conclusions:

• Supervised ML effectively identifies genes associated with persistent MODS.
• This approach holds potential for informing targeted interventions in at-risk populations, pending further validation in diverse cohorts.