AI MARRVEL: A new AI tool to diagnose genetic disorders

Identifying the genes responsible for rare genetic disorders is labor-intensive and requires intricate decision-making and existing computational tools have so far achieved moderate success in helping geneticists accelerate this process. A new tool called AI-MARRVEL (AIM) developed by researchers at the Jan and Dan Duncan Neurological Research Institute (Duncan NRI) at Texas Children's Hospital and Baylor College of Medicine is a major stride forward in the use of artificial intelligence (AI) to guide the discovery of previously undiagnosed genetic disorders. 

Dr. Zhandong Liu, Chief Computational Scientist at Texas Children’s and associate professor at Baylor College of Medicine, led the team that developed this tool. A paper describing its development was published in the NEJM AI.

To diagnose ambiguous disease presentations, physicians establish a differential diagnosis, carefully review the possible causes, and choose the most appropriate test for confirmation. However, this is far more challenging for disorders that are not well-characterized and often are associated with rare genetic variants. 

Millions of individuals all over the world are affected by rare genetic disorders that are caused by one or more variants in one gene. However, every individual’s genome has tens of thousands of variants when compared to the reference genomes. Yet, despite using many sophisticated bioinformatics tools to remove common variants that are less likely to contribute to a disorder, clinical geneticists are often left with hundreds of potential gene candidates. Sifting through these candidates to identify the correct gene variant is a time-consuming and costly process, requiring expert knowledge of various fields of biology and bioinformatics. Using current approaches, the diagnostic rate for previously undiagnosed individuals is only 30-40%, which leaves many individuals without a correct molecular diagnosis.

“We recognized the critical need for efficient, systematic, and comprehensive approaches to enhance the accuracy and speed of diagnosis for rare and undiagnosed conditions,” Dr. Liu said. “Our goal was to accelerate the discovery of new disease genes and improve prediction accuracy by creating a knowledge- and AI-driven system that can take into account the patient’s clinical symptoms, sequencing results, and prioritize candidate gene variants based on foundational genetics principles and key decision-making insights from clinical geneticists to accurately identify the most likely causative gene(s).”

A few years back, Dr. Liu’s team created MARRVEL, a publicly available website that integrates numerous human and model organism gene-centric resources to facilitate the analysis of the human gene and variants. To create AIM, they improved upon that tool by training it with high-quality samples that were clinically diagnosed and curated by American Board of Medical Genetics and Genomics (ABMGG) board-certified experts. Further, they trained this software to learn the complex decision-making logic involved in molecular diagnosis, using additional expert-engineered features that encode prior knowledge such as fundamentals of genetics and the clinical expertise of geneticists. AIM can greatly impact the way genetic disorders are diagnosed by improving the efficiency and reducing the workload of diagnosis by reducing potential differential diagnoses and/or potential gene candidates and by allowing quick and affordable large-scale, automatic re-analysis of undiagnosed cases.

“AIM achieved a precision of 98%, identified 57% of diagnosed cases, and outperformed all benchmarked methods when tested on three independent databases,” Dr. Liu said. “We are excited by the significant advances this tool can bring about to the field of clinical genetics diagnostics by increasing the speed and accuracy of the predictions while reducing the cost of novel disease gene discovery,”

Other authors of this work include Dongxue Mao, Chaozhong Liu, Linhua Wang, Rami AI-Ouran, Cole Deisseroth, Sasidhar Pasupuleti, Seon Young Kim, Lucian Li, Jill A.Rosenfeld, Linyan Meng, Lindsay C. Burrage, Michael Wangler, Shinya Yamamoto, Michael Santana, Victor Perez, Priyank Shukla, Christine Eng, Brendan Lee and Bo Yuan. They are affiliated with one or more of the following institutions: Baylor College of Medicine, Jan and Dan Duncan Neurological Research Institute at Texas Children’s Hospital, Al Hussein Technical University, Baylor Genetics and the Human Genome Sequencing Center at Baylor. This work was supported by the NIH grants, Chan Zuckerberg Initiative, the Chao Endowment, the Huffington Foundation and the Jan and Dan Duncan Neurological Research Institute at Texas Children's Hospital. 

Read Baylor's news release here.