Machine Learning-Assisted Multisensory Platform Based on Process-Engineered Reduced Graphene Oxide Transistors to Detect Autism-Related Metabolites

Abstract Autism Spectrum Disorder (ASD) is a heterogeneous neurodevelopmental condition primarily diagnosed through behavioral assessments. Despite rising diagnosis rates, behavioral analysis can overlook subtle cases, highlighting the need for complementary sensing technologies. However, the absence of definitive ASD biomarkers and receptors hinders traditional analytical sensor development. While altered urinary levels of gut microbiota-derived metabolites, including 3-(3-hydroxyphenyl)-3-hydroxypropionic acid (HPHPA) and 3-hydroxyhippuric acid (3HHA), are reported in autistic individuals, their undefined concentration thresholds are also a challenge. To overcome this, we present a machine learning-driven multisensory platform based on electrolyte-gated transistors (EGTs) fabricated from reduced graphene oxide (rGO) with varying properties (i.e., surface roughness, defects, carbon/oxygen ratio, etc.). The process-engineered rGO produces nonspecific EGT sensors that generate a cross-sensitive electrical fingerprint capable of detecting the target analytes at a wide range of concentrations in artificial urine. The platform demonstrates high sensitivity, with regression models achieving strong determination coefficients (R2) of 0.989 for HPHPA and 0.996 for 3HHA. Advanced data analysis was also used to investigate which EGT parameters contribute to the sensing response. Our results highlight the potential of pattern-based multisensory platforms for detecting small-molecule analytes in complex media, including ASD-related metabolites, ultimately advancing electronic sensors for translational research.