Phenotype driven molecular genetic test recommendation for diagnosing pediatric rare disorders

Area of Science:

• Genetics and Genomics
• Machine Learning in Medicine
• Rare Disease Diagnostics

Background:

• Rare disease patients face significant diagnostic delays.
• Genetic testing is vital but complex for non-specialists.
• Current guidelines suggest exome/genome sequencing or gene panels based on clinical presentation.

Purpose of the Study:

• To develop a machine learning model for recommending appropriate genetic tests.
• To assist general pediatricians in navigating complex genetic testing decisions.
• To expedite the diagnosis of rare diseases.

Main Methods:

• A machine learning model was trained on 1005 patient records from Columbia University Irving Medical Center.
• The model utilizes patient phenotype information to recommend genetic tests.
• Performance was evaluated using Area Under the Receiver Operating Characteristic Curve (AUROC) and Area Under the Precision-Recall Curve (AUPRC).

Main Results:

• The model achieved an AUROC of 0.823 and AUPRC of 0.918 on the training cohort.
• The model demonstrated strong generalizability in an external cohort with AUROC:0.77 and AUPRC: 0.816.
• Model performance closely aligned with decisions made by genetic specialists.

Conclusions:

• The developed machine learning model effectively recommends genetic tests for rare disease diagnosis.
• The tool shows potential to aid general pediatricians in improving diagnostic timelines.
• This approach can enhance genetic test ordering and expedite rare disease identification.