Machine learning classification of ADHD and HC by multimodal serotonergic data.

Area of Science:

• Neuroscience
• Genetics
• Psychiatry

Background:

• Serotonin neurotransmission is implicated in attention-deficit/hyperactivity disorder (ADHD) etiology and pathology.
• Single nucleotide polymorphisms (SNPs) in serotonin-related genes may mediate these effects.

Purpose of the Study:

• To develop and validate a multivariate classification model for ADHD using genetic data and positron emission tomography (PET) imaging.
• To identify key genetic markers and brain regions discriminative of ADHD.

Main Methods:

• PET imaging with [11C]DASB to measure serotonin transporter (SERT) binding potential in 16 ADHD patients and 22 healthy controls (HC).
• Genotyping for 30 SNPs in HTR1A, HTR1B, HTR2A, and TPH2 genes.
• Random forest machine learning with five-fold cross-validation for feature selection and classification.

Main Results:

• The model identified specific SNPs (HTR2A rs1328684, rs6311; HTR1B rs130058) and brain regions (posterior cingulate gyrus, cuneus, precuneus, pre/para/postcentral gyri) as most discriminative.
• Achieved a mean prediction accuracy of 0.82 (±0.09) with balanced sensitivity (0.75) and specificity (0.86).

Conclusions:

• The findings support the relevance of SERT, HTR1B, and HTR2A genes in ADHD.
• The developed model shows potential as a computer-aided diagnostic tool for ADHD, aiding clinical decisions, especially given high comorbidity rates.