A pilot study on machine learning approach to delineate metabolic signatures in intellectual disability.

Intellectual disability (ID) is a neurodevelopmental disorder characterized by cognitive delays. Inborn errors of metabolism constitute an important subgroup of ID for which various treatments options are available. We aimed to identify potential biomarkers of inherited metabolic disorders from the children with ID using tandem mass spectrometry and develop a novel machine learning algorithm to differentiate between the cases and the controls. All of the cases were having IQ score <70, gross motor delay, speech disorder and no recognizable symptoms of the condition. Metabolite profiling of ID individuals exhibited low tyrosine/large neutral amino acids, high citrulline/arginine ratios; elevated proline, alanine, phenylalanine, and ornithine, while a significant decrease in the level of amino acid arginine, and elevated C4 (butyrylcarnitine) and C4OH/C3DC (3-hydroxybutyrylcarnitine/malonylcarnitine). Machine learning algorithm differentiated cases and controls efficiently using specific thresholds of ornithine, arginine and C4OH/C3DC. Furthermore, ID cases were distinguished into mild, moderate, and severe based on specific thresholds of methionine, arginine, and C5OH/C4DC (3-hydroxyisovalerylcarnitine/methylmalonylcarnitine). The machine learning algorithm could successfully identify specific metabolite markers in ID and correlate the same with neurological features.