Recently, researchers used Fabric GEM, an artificial intelligence algorithm, to detect disease causing variants in newborns and rare disease patients with over 90% accuracy.
Around 7 million infants are born with serious genetic disorders worldwide each year. During the last decade, with increased knowledge of genes associated with Mendelian conditions and major improvements in whole genome sequencing (WGS) and whole exome sequencing (WES) rates of diagnosis have increased from about 10% to over 50%.
However, barriers still remain surrounding the clinical elucidation of genetic variants. This is largely due to clinical genome interpretation being a very manual and labour-intensive process. Moreover, the practise needs to be performed by highly trained genome analysts, genetic counsellors and laboratory directors. On average, if 100 variants are reviewed per case, this translates into up to 100 hours of expert time per patient. In practise, this has led to the review of only 10 variants per case, which somewhat defeats the purpose of genome-wide sequencing. This has ultimately resulted in a lack of widespread adoption of genomic testing for patients with suspected genetic disorders.
Using Fabric GEM to improve genome interpretation
AI has already made significant progress in healthcare, for example in assisting drug discovery and enhancing precision medicine. Now it holds promise to greatly improve genome interpretation by integrating predictive methods with the growing knowledge of genetic diseases. Recently, researchers used an AI algorithm called Fabric GEM, developed by Fabric Genomics, to detect disease-causing variants in newborns and rare disease patients. The results were published as a retrospective study in Genome Medicine.
Fabric GEM is a new AI-based electronic clinical decision support system that can be used to enhance genome interpretation. First, the researchers assessed the diagnostic performance of Fabric GEM by applying it to a diverse cohort of retrospective paediatric cases from Rady Children’s Institute for Genomic Medicine. These cases were mostly seriously ill infants, all of whom had been diagnosed with Mendelian conditions following WGS or WES. Then, the researchers replicated these analyses in a separate cohort of 60 cases, collected from five additional medical centres. Patient phenotypes were also extracted from clinical notes both manually and using an automated clinical natural language processing (CNLP) tool to compare with current variant prioritisation tools.
It was found that Fabric GEM correctly identified over 90% of the top causal genes and it prioritised an average of three candidate genes for review per case. Moreover, the AI algorithm was able to rank specific diseases associated with these genes to assist clinicians in the diagnosis of each case. Ultimately, despite the differences in case collection, sequencing methods and bioinformatic pipelines across all sites, Fabric GEM’s performance demonstrated a new standard of accuracy.
Possibilities of AI in genomic interpretation
These findings demonstrate that rapid, accurate and comprehensive genomic-based diagnosis is achievable through innovations made possible by AI. The researchers showed that Fabric GEM could be used to successfully reduce the burden of gene variant review by clinical geneticists, in turn reducing the obstacles facing the adoption of genomic testing.
Stephen Kingsmore, the President and CEO of Rady Children’s Institute for Genomic Medicine, explained: “Fast and definitive genetic diagnosis is essential to providing the right treatment in a timely manner for critically ill new-borns. Fabric GEM has successfully demonstrated that it can automatically and quickly suggest a very short list of candidate genes for interpretation through whole-genome or whole-exome sequencing.”
Check out our exclusive interview with co-author of this study, Stephen Kingsmore, as he discusses his early career, his move to Rady Children’s Institute for Genomic Medicine and his Guinness World Record.
Image credit: Forbes
