AI-MARRVEL - A Knowledge-Driven AI System for Diagnosing Mendelian Disorders

Area of Science:

• Genetics
• Bioinformatics
• Artificial Intelligence

Background:

• Genetic disorder diagnosis is complex, requiring extensive manual interpretation of variants.
• Current artificial intelligence (AI) tools offer moderate success in primary genetic diagnoses.

Purpose of the Study:

• To develop and evaluate AI-MARRVEL (AIM), an AI tool to enhance genetic disorder diagnosis.
• To improve the accuracy and efficiency of identifying disease-causing genetic variants.

Main Methods:

• AI-MARRVEL (AIM) utilizes a random-forest machine-learning classifier trained on over 3.5 million variants.
• Incorporates expert-engineered features to mimic human diagnostic decision-making.
• Evaluated on three independent patient cohorts.

Main Results:

• AIM doubled the rate of accurate genetic diagnoses compared to benchmark methods across cohorts.
• A confidence metric achieved 98% precision, identifying 57% of diagnosable cases from unsolved pools.
• AIM showed potential for novel disease gene discovery, predicting two new disease genes.

Conclusions:

• AIM demonstrates superior accuracy for genetic diagnosis compared to existing methods.
• The tool can aid in primary diagnosis, reanalysis of unsolved cases, and novel gene discovery.
• AIM has the potential to accelerate genetic research and clinical applications.