AI improves the diagnosis of rare genetic disorders

Diagnosing rare Mendelian disorders is labor-intensive, even for experienced geneticists. Researchers at Baylor College of Medicine are working to make this process more efficient with the help of artificial intelligence. They have developed a machine learning system AI-MARRVEL (AIM) that helps prioritize potentially causative variants of Mendelian disorders. The study was published in 'The New England Journal of Medicine'

The Baylor Genetics researchers found that AIM can make predictions independent of clinical knowledge, driving the discovery of new disease mechanisms. "The diagnosis rate for rare genetic disorders is only about 30%, and on average it takes six years from symptom onset to diagnosis. There is an urgent need for new approaches to improve the speed and accuracy of diagnosis," said Dr. Pengfei Liu, co-author of the study, associate professor of molecular and human genetics and associate clinical director at Baylor Genetics.

AI-MARRVEL offers new hope for cases of rare diseases that have remained unsolved for years. Hundreds of new disease-causing variants are reported each year that could be critical to solving these unsolved cases. However, determining which cases warrant reanalysis is challenging due to the large number of cases. The researchers tested the clinical exome reanalysis of AIM on UDN and DDD case datasets and found that it could correctly identify 57% of diagnosable cases.

AIM is trained against a public database of known variants and genetic analyzes called Model Organism Aggregated Resources for Rare Variant Exploration (MARRVEL), which was previously developed by the Baylor team. The MARRVEL database contains over 3.5 million variants from thousands of diagnosed cases. The researchers provide AIM with the patients' exome sequence data and symptoms, and AIM ranks the most likely candidate genes causing the rare disease.

The researchers compared AIM's results with other algorithms used in recent benchmark articles. They tested the models using three cohorts of data with established diagnoses from Baylor Genetics, the Undiagnosed Diseases Network (UDN) funded by the National Institutes of Health, and the Deciphering Developmental Disorders (DDD) project. AIM consistently ranked the diagnosed genes as the top candidates in twice as many cases as all other benchmark methods using these real-world datasets.

"We trained AIM to mimic the way humans make decisions, and the machine can do this much faster, more efficiently and more cost-effectively. This method effectively doubled the rate of correct diagnoses," said Dr. Zhandong Liu, co-author of the study, associate professor of pediatrics and neurology at Baylor College and a researcher at the Jan and Dan Duncan Neurological Research Institute (NRI) at Texas Children's Hospital.

"We can make the process of reanalysis much more efficient by using AIM to identify a set of very confident, potentially solvable cases and advancing these cases for manual review," Zhandong Liu explains. "We anticipate that this tool can uncover an unprecedented number of cases that were previously considered undiagnosable."

The researchers also tested the potential of AIM to discover new candidate genes not previously associated with disease. AIM correctly predicted two newly reported disease genes as top candidates in two UDN cases.

"AIM is a significant step forward in the use of AI to diagnose rare diseases. It reduces genetic differential diagnoses to a few genes and has the potential to advance the discovery of previously unknown diseases," comments Dr. Hugo Bellen, co-author of the study, Distinguished Service Professor of Molecular and Human Genetics at Baylor College and Chair of Neurogenetics at Duncan NRI.

"Combined with the extensive expertise of our certified clinical laboratory directors, highly curated data sets and scalable, automated technology, we are seeing the impact of augmented intelligence to deliver comprehensive genetic insights at scale, even for the most vulnerable and complex patient populations," said Dr. Fan Xia, lead author, associate professor of molecular and human genetics at Baylor and vice president of clinical genomics at Baylor Genetics.

"By applying real-world training data from a Baylor Genetics cohort without any inclusion criteria, AIM has demonstrated superior accuracy. Baylor Genetics is committed to developing the next generation of diagnostic intelligence and bringing it into clinical practice," concluded Dr. Fan Xia.