Identification of biological mechanisms underlying a multidimensional ASD phenotype using machine learning.

Area of Science:

• Genetics
• Neuroscience
• Computational Biology

Background:

• Autism Spectrum Disorder (ASD) presents complex genetic and phenotypic heterogeneity, complicating molecular diagnosis and prognosis.
• Understanding genotype-phenotype correlations is crucial for advancing ASD research and clinical applications.

Purpose of the Study:

• To develop an integrative machine-learning approach for precise genotype-phenotype correlations in ASD.
• To identify distinct ASD phenotypic subgroups and their associated biological processes based on copy number variants (CNVs).

Main Methods:

• Utilized clustering analysis on clinical data from 2446 ASD cases to identify phenotypic subgroups.
• Performed functional enrichment analysis of brain genes affected by CNVs.
• Developed a Naive Bayes classifier to predict phenotypic clusters from disrupted biological processes.

Main Results:

• Identified two distinct ASD phenotypic subgroups differing in severity, adaptive behavior, intellectual ability, and verbal status.
• Found 15 significant biological processes disrupted by CNVs, including neural development and cognition.
• The classifier achieved high precision (0.82) but low recall (0.39) in predicting clusters, dependent on patient information content.

Conclusions:

• Milder and more severe ASD clinical presentations may involve distinct biological mechanisms.
• Machine learning can reduce clinical heterogeneity by integrating multidimensional data, establishing genotype-phenotype correlations.
• Findings represent a step towards translating genetic information into clinical applications, highlighting the need for comprehensive datasets.