Diagnosing rare Mendelian disorders is a challenging task that typically involves labor-intensive efforts from geneticists. To address this issue, researchers at Baylor College of Medicine have developed a machine learning system called AI-MARRVEL (AIM) to help prioritize potentially causative variants for these disorders. By leveraging a public database of known variants and genetic analysis called Model organism Aggregated Resources for Rare Variant ExpLoration (MARRVEL), AIM aims to enhance the speed and accuracy of diagnosis for rare genetic disorders, which currently have a diagnostic rate of only about 30%.
The AIM system was trained using the MARRVEL database, which contains millions of variants from thousands of diagnosed cases. Researchers provide AIM with patients’ exome sequence data and symptoms, and AIM then ranks the most likely gene candidates causing the rare disease. When compared to other algorithms used in recent benchmark papers, AIM consistently ranked diagnosed genes as the top candidate in twice as many cases as other methods across three data cohorts from Baylor Genetics, the Undiagnosed Diseases Network (UDN), and the Deciphering Developmental Disorders (DDD) project.
The researchers behind AIM believe that the system could significantly increase the rate of accurate diagnosis for rare diseases. They trained AIM to mimic the decision-making process of humans, resulting in a machine that can make decisions faster, more efficiently, and at a lower cost. AIM also offers hope for unsolved rare disease cases that have remained unresolved for years, as it was able to correctly identify 57% of diagnosable cases in a dataset of UDN and DDD cases. By identifying a high-confidence set of potentially solvable cases for manual review, AIM could potentially recover many cases that were previously thought to be undiagnosable.
In addition to its diagnostic capabilities, AIM also has the potential to contribute to the discovery of novel disease genes. The system correctly predicted two newly reported disease genes as top candidates in two UDN cases, demonstrating its ability to identify gene candidates that have not been linked to a disease before. According to the researchers, AIM represents a significant step forward in using AI to diagnose rare diseases, narrowing down genetic diagnoses to a few key genes and potentially guiding the discovery of previously unknown disorders.
The researchers involved in the development of AIM, including Dr. Hugo Bellen, Dr. Fan Xia, and others, believe that the combination of AI technology, curated datasets, and expert clinical knowledge has the potential to provide comprehensive genetic insights at scale, even for complex conditions. By applying real-world training data from a Baylor Genetics cohort, AIM has shown superior accuracy and aims to become the next generation of diagnostic intelligence for clinical practice. This work was supported by the Chang Zuckerberg Initiative and the National Institute of Neurological Disorders and Stroke.
