An investigation utilizing machine learning (Recursive Ensemble Feature Selection, REFS) identifies potential biomarkers for asthma control in pediatric patients.
A potential link between asthma control and the gastrointestinal microbiome was observed in a study suggesting a new useful biomarker for identifying children with uncontrolled asthma. With machine learning, findings identified microorganisms that could differentiate children with controlled and uncontrolled asthma: haemophilus and veillonella.
Global diversity and conventional differential abundance analysis (DAA) did not exhibit significant differences between the groups of pediatric patients with asthma. However, machine learning (Recursive Ensemble Feature Selection, REFS) detected a set of taxa, including haemophilus and veillonella, which were able to differentiate between uncontrolled and controlled asthma with an average classification accuracy of 81% (saliva) and 86% (feces).
According to the study, these taxa were enriched in taxa previously associated with inflammatory diseases for both sampling compartments, and with chronic obstructive pulmonary disease (COPD) for the saliva samples.
This research presents opportunities for for further investigation into machine learning and the differentiation of controlled and uncontrolled children with asthma based on their gastrointestinal microbiome. With the potential of a new biomarker for treatment responsiveness, investigators wrote that these data could help to improve asthma control in children.
As a chronic respiratory disease that affects millions of people worldwide, uncontrolled asthma can lead to severe exacerbations and reduced quality of life. Previous research has linked the microbiome with asthma characteristics, but this investigation specifically targeted the association with asthma control, which had not been explored until now, according to the study.
A team of investigators led by Jelle Blankestijn, PhD candidate, Department of Pulmonary Medicine, Amsterdam UMC, University of Amsterdam, conducted the analysis on data collected from 143 children with moderate-to-severe asthma aged 6-17 years. A total of 246 fecal and saliva samples were obtained as part of the SysPharmPediA study.
Patients were classified with either controlled or uncontrolled asthma based on their asthma symptoms, and their microbiome at the species level was compared using global diversity, conventional differential abundance analysis, and machine learning (Recursive Ensemble Feature Selection, REFS).
The study has several limitations, including its small sample size and that the study only focused on children with moderate-to-severe asthma. Investigators stated the need for further research to confirm these findings in larger and more diverse populations.
However, the importance of the data was acknowledged, as it illuminated the
possibilities for identifying biomarkers provided by machine learning, specifically in complex diseases like asthma and can inform future research.