Machine learning based analysis for intellectual disability in Down syndrome.

Area of Science:

• Genetics and Bioinformatics
• Artificial Intelligence in Medicine

Background:

• Down syndrome (DS), or trisomy 21, is the leading genetic cause of intellectual disability (ID), yet its pathogenic mechanisms remain unclear.
• Traditional analysis methods struggle with complex conditions like DS, necessitating advanced approaches for identifying cause-effect relationships.

Purpose of the Study:

• To apply machine learning (ML) techniques to analyze clinical records of individuals with DS and identify features associated with intellectual disability.
• To investigate the correlation between various clinical features and intellectual functioning in DS using ML models.

Main Methods:

• Utilized two tree-based ML models: random forest and gradient boosting machine.
• Analyzed 109 features across 106 DS subjects, using age equivalent (AE) score as the indicator of intellectual functioning.
• Implemented Boruta feature selection, data augmentation, and age effect mitigation to refine model accuracy and reliability.

Main Results:

• ML algorithms demonstrated good accuracy in identifying variables linked to cognitive impairment in DS.
• Random forest and gradient boosting machine models achieved low error rates (MSE <0.12) and acceptable R² values (0.70 and 0.93).
• Key features associated with ID in DS included hearing, gastrointestinal issues, thyroid function, immune status, and vitamin B12 levels.

Conclusions:

• ML models can effectively identify features correlated with ID in DS, aiding research into therapeutic targets and care pathways.
• This study provides a foundation for further validation of ML algorithms with larger datasets to enhance understanding and management of DS.