Symptom-based classification of 16p11.2 copy number variations underlying the multidimensional autism spectrum disorder phenotype using machine learning methods
Research in Autism, cilt.132, 2026 (SSCI, Scopus)
- Yayın Türü: Makale / Tam Makale
- Cilt numarası: 132
- Basım Tarihi: 2026
- Doi Numarası: 10.1016/j.reia.2026.202865
- Dergi Adı: Research in Autism
- Derginin Tarandığı İndeksler: Social Sciences Citation Index (SSCI), Scopus
- Anahtar Kelimeler: 16p11.2, CNVs, Machine Learning, Neurodevelopmental Disorders
- Sağlık Bilimleri Üniversitesi Adresli: Evet
Özet
Purpose: Copy number variations (CNVs) in the 16p11.2 region are well-established contributors to neurodevelopmental disorders, yet phenotype variability across this locus remains insufficiently characterized. This study investigates clinical features and ASD-related symptoms among carriers of rare pathogenic and common CNVs, and evaluates symptom-level discriminability using machine learning (ML) methods. Methods: Genetic data from 7568 individuals were retrospectively screened, identifying 147 carriers of 16p11.2 CNVs. Detailed clinical assessments were completed for 50 participants. ASD-related symptoms were evaluated using a structured 25-item instrument. Group comparisons applied nonparametric statistics with effect sizes, confidence intervals, and FDR correction. ML analyses used PCA and k-means for feature selection, oversampling methods (SMOTE, Borderline-SMOTE, ADASYN), and five classifiers, evaluated through cross-validation. Results: Across pathogenic and common CNV groups, no significant differences were observed in social communication, restricted/repetitive behaviors, sensory symptoms, regression, or total autism scores (FDR-adjusted p > 0.05). Aggression was more frequently endorsed in pathogenic CNV carriers (raw p = 0.030; FDR p = 0.098). BMI was higher in pathogenic CNVs, though nonsignificant after correction (raw p = 0.027; FDR p = 0.152). ML analyses identified three recurrent discriminative symptoms across multiple datasets: delayed response to name, unusual object play, and aggression. Dataset 3 (16 symptoms) provided the most balanced classification performance but, given the very small pathogenic CNV sample, results remain exploratory. Conclusion: Findings suggest that, while most autism-related symptoms do not differ between groups, aggression and increased BMI may represent preliminary phenotypic signals associated with pathogenic CNVs. Integrating clinical data from 147 CNV carriers further supports a potential widespread effect across the broader 16p11.2 locus rather than a single breakpoint-specific mechanism. However, all results should be interpreted cautiously due to limited sample size, and larger, systematically phenotyped cohorts are required to establish robust genotype–phenotype relationships.