Phenotype based prediction of exome sequencing outcome using machine learning for neurodevelopmental disorders

Area of Science:

• Genetics
• Machine Learning
• Clinical Diagnostics

Background:

• Exome sequencing (ES) has improved neurodevelopmental disorder (NDD) diagnosis but has a stable diagnostic yield of ~30% in clinical practice.
• Improving patient selection based on phenotype is crucial to enhance diagnostic yield and minimize unnecessary genetic testing.

Purpose of the Study:

• To develop a predictive model, PredWES, that estimates the probability of a positive ES result based on patient phenotype.
• To enhance the diagnostic yield of ES for patients with NDDs through phenotype-driven selection.

Main Methods:

• Tested four machine learning methods to develop PredWES, a statistical model for predicting ES results.
• Trained the model on 1663 patients with NDDs, focusing on phenotypic presentation.

Main Results:

• Applying PredWES to the top 10% of patients predicted to have a positive ES result achieved a 56% diagnostic yield, an 114% increase.
• Phenotypic features like low muscle tone and microcephaly correlated positively with conclusive ES diagnoses in NDD patients.
• Autism diagnosis showed a negative correlation with achieving a molecular diagnosis via ES.

Conclusions:

• PredWES effectively prioritizes NDD patients for diagnostic ES based on phenotypic data.
• This approach significantly increases diagnostic yield and optimizes the use of healthcare resources.
• Phenotype-based prediction models can improve the efficiency of genetic testing in clinical settings.