NEW YORK – Two recent studies lend support to Fabric Genomics' artificial intelligence-based algorithm to help diagnose genetic diseases, providing further validation of the end-to-end software tool intended to help labs launch genomic testing programs.
Launched in October 2020, the Fabric GEM algorithm was developed in collaboration with the University of Utah and is designed to draw on next-generation sequencing data and accelerate diagnosis of a patient's condition.
In one study, a retrospective study in Genome Medicine published this past October, researchers from Rady Children's Institute for Genomic Medicine used the algorithm to analyze data from 119 patients across six genomic centers and hospitals. The researchers were able to detect more than 90 percent of disease-causing variants in newborns and rare disease patients using the algorithm. Beyond the variants, the tool also ranked specific diseases and conditions associated with the genes to help clinicians diagnose each case.
And in a more recent study published in the New England Journal of Medicine earlier this month, Fabric's technology was used with other tools in the pilot stage of the 100,000 Genomes Project to demonstrate that whole-genome sequencing can help with diagnosis of a significant subset of rare disease cases.
Researchers from a variety of sites, including Genomics England, National Health Service England, and Queen Mary University of London, analyzed data from 4,660 participants from rare disease-affected families who had their genomes sequenced by Illumina. Fabric Genomics and Congenica contributed to the phenotype-based variant prioritization steps to find the most promising genetic variants.
Using the tools, the team was able to successfully diagnose 25 percent of cases and discover new and known pathogenic or likely pathogenic mutations. For patients with hearing or vision, metabolic, neurological, or intellectual disability symptoms, researchers were able to diagnose between 40 percent and 55 percent of diseases.
Fabric Genomics CEO Martin Reese was careful to note that GEM isn't an actual test but rather a tool that can act as a "plug and play" for any lab looking to implement genomic testing. The company overall, he added, is not a diagnostic firm.
Beyond the full software suite, called Fabric Enterprise, the company also offers an interpretation team that can analyze and explain the data if labs don't want to be responsible for interpretation themselves.
Fabric Enterprise includes GEM, the phenotype-matching algorithm for genomic exomes and genomes, and Fabric ACE, a separate algorithm for genomic gene panels. For the genome analysis conducted with Rady, GEM was coupled with partner company Clinithink's CLiX focus, a natural language processing technology that can sort through medical notes and pick out key terms to aid in diagnosis.
GEM uses the whole-genome or whole-exosome sequencing data to produce a list of the most likely variants causing a patient's genetic condition, with each variant scored based on likelihood. That process can take up to 30 minutes, although Reese said there are times when GEM hits on a result immediately.
Meantime, CLiX, which can be bought through Fabric Enterprise as an add-on, uses electronic health record info to more fully support an official diagnosis. One bottleneck for results interpretation is that patients often have a lot of medical record data that, in traditional interpretation, would need to be weeded through by a clinical geneticist.
CLiX automatically extracts key terms from the notes that can signify a specific disease, including ICD codes and human phenotype ontology terms indicating symptoms or diseases, and matches them with a library of more than 12,000 clinical characteristics of rare diseases, Reese said.
The normal extraction process can take up to six hours of hands-on time with a geneticist, and manual interpretation has a limited number of terms — around six — that can be looked for, due to time constraints, Reese said. But CLiX can identify 100 to 200 terms at a time, providing a more comprehensive overview and, when used with GEM, more specific diagnosis, Reese said.
Once the data from GEM and CLiX is combined and a final report is generated, the clinical lab director reviews the report and signs off on it, so there's an extra level of verification before the official diagnosis, Reese added.
Although Fabric has a manual interpretation team that can analyze the final report from Fabric Enterprise if labs opt in, Reese said laboratories are also welcome to have geneticists interpret the results on their own. He was also careful to note that Fabric does not perform any of the testing itself but does all of the processing and interpretation of that testing.
Nearly 100 hospital and clinical laboratory customers are using the Fabric Enterprise software suite, Reese said. The company, which has raised $32 million in funding, is hoping to capitalize on the two recent GEM studies to build its customer base.
Mark Yandell, a professor of human genetics at the University of Utah and a founding scientific adviser to Fabric who wrote the GEM algorithm, said that a key problem with genetic interpretation is that it takes time, effort, and money to review all of the information at a clinician's disposal. But an algorithm like GEM, and Fabric Enterprise overall, can significantly reduce the time and effort, as well as the cost, he said.
Yandell noted that the tool can compare the information gleaned from the patient's genome sequence and medical notes with online databases of genetic diseases and return results in approximately four minutes.
Stephen Kingsmore, president and CEO of Rady and a co-author of the Genome Medicine study who has used Fabric's interpretation software for the past five years, also noted the timesaving potential of Fabric's software for lab directors. He said via email that GEM doesn't replace manual interpretation but is a "great addition" and a good way to begin interpreting a patient's case. Rady currently uses GEM as part of its routine approach to interpretation of rapid diagnostic whole-genome sequencing.
"High scoring variants" from GEM are "well worth checking into," and it can also be used as "one last check" to ensure nothing's been missed as a clinician is finishing up a case, Kingsmore said.
He also noted that the platform could be used as an initial implementation of genetic testing in situations where there is "extreme resource limitation."
However, he added that the algorithm doesn't yet extend to structural and copy number variants, but that Fabric is continuing to develop and optimize the tool.
The platform can be customized for other kinds of NGS-based assays, although the company's focus is on "highly reimbursed and highly beneficial tests," Reese said. There are so many clinical molecular tests that use NGS that Fabric can be applied to "any kind of clinically relevant test," he added.
Its pricing is based on the test being performed, and the firm usually receives a portion of the test's total reimbursement, although that depends on what specific services a lab is using and whether it's drawing on Fabric's interpretation team or using its own, Reese said.