Using machine learning, a field closely related to artificial intelligence, upon nuclear magnetic resonance (NMR) spectra of exhaled breath condensate, Delhi-based researchers have been able to improve the diagnosis of childhood asthma and even identify three asthma subtypes. This pushes the current understanding of childhood asthma towards having metabolomic (study of chemical processes involving metabolites) subtypes, which have been largely unknown so far.
A team of researchers led by Dr. Anurag Agrawal from Delhi’s CSIR-Institute of Genomics and Integrated Biology (IGIB) and Dr. Tavpritesh Sethi from IIIT-Delhi and AIIMS has now achieved a measure of success. The researchers have been able to correctly identify children with asthma and also the subtypes along with potential biomarkers.
The study included 89 asthmatic children below 18 years and 20 healthy individuals with no history or clinical manifestation of asthma to identify the NMR signatures of asthmatic children; the NMR spectra of 61 asthmatic children with clinical data were used for identifying the subtypes. In an ongoing cohort at AIIMS, the children have been followed up for five years now, says Dr. Koundinya Desiraju from CSIR-IGIB and one of the first authors of the paper published in the Journal of Translational Medicine.
Unlike other researchers who looked for specific metabolites in exhaled breath using NMR, Dr. Agrawal and Dr. Sethi looked for global NMR signatures of all metabolites from exhaled breath that was condensed at -80 degree C. “Unknown and highly variable dilution of exhaled breath has been a major problem in this field. Unlike in the case when specific metabolites are looked for, the overall shape of the signature will remain the same immaterial of the dilution of exhaled breath,” says Dr. Agrawal.
But the challenge with studying global signatures is that human eye is not equipped to seeing hundreds of peaks and picking out a pattern. This is where artificial intelligence came in handy. The algorithm was able to differentiate the total NMR spectrum (which was normalised) of healthy children and those who had asthma. It could also identify three subtypes of asthma. The algorithm has 80% sensitivity and 75% specificity in identifying children with asthma.
“We could correlate the different subtypes with different clinical manifestations,” says Dr. Agrawal.
Children belonging to subtype 1 showed a typical signature of ammonia metabolite but had no family history of asthma. “This asthma subtype is more like the typical allergic form of asthma,” says Dr. Sethi. “But subtype 3 had lower blood eosinophilia and elevated neutrophilia compared with the other two subtypes. Children belonging to this subtype had a stronger family history of asthma and suffered from more acute asthma episodes even when on treatment.” Subtype 3 showed a peak corresponding to formic acid.
Subtype 2 had high eosinophil count but was otherwise similar to subtype 1, but very different from subtype 3. “Not every chemical difference in exhaled breath will translate into clinical difference. To know if there is any clinical difference in children belonging to subtype 1 and 2, more children have to be followed up for a longer period of time,” Dr. Agrawal says.
Currently, there is no difference in treatment for children belonging to three subtypes. “Knowing the difference between subtype 1 and 2 as one group and subtype 3 will help in better treatment strategies, which is the goal of precision medicine,” Dr. Sethi says. “Children with subtype 3 asthma may need more aggressive therapy or alternative treatment strategy. But at this stage of the study we don’t know the details.”
The next step will be to validate the signatures as biomarkers of asthma subtypes. For this, the subtype 1 and 2 will be looked together and contrasted with subtype 3.
VISIT THE SOURCE ARTICLE