NEW YORK – It may take a year, or even two, but rare disease patients are increasingly getting genetic diagnoses thanks to software that is turning genome analysis into a continuous process rather than a one-off event.
Scanning the ever-increasing pile of gene-disease associations and updates to pathogenicity classifications of genetic variants, new programs are connecting old patient data to new findings, often automatically. Variations on this theme are getting put into place in leading rare disease clinical labs in the US and Australia.
At the Broad Institute's Center for Mendelian Genetics, researchers are using the approach to aid in discovering new gene-disease relationships, and Ambry Genetics has automated genomic data reanalysis as part of its exome testing service. Clinicians at Australia's Centre for Population Genomics, a joint initiative of the Garvan Institute of Medical Research and the Murdoch Children's Research Institute, are putting together a program to evaluate the feasibility of a nationwide network to constantly reanalyze patient genomes.
Both the Broad and Australian efforts rely on automated software developed in collaboration with Microsoft to help them reanalyze rare disease patient genomes.
The Australian team is far from the first group to pursue this goal, but it's built software that offers "close to continuous" monitoring of new findings while also being able to scale across thousands of datasets, according to Zornitza Stark, a clinical geneticist at Murdoch Children's Research Institute who leads the project. Moreover, they've begun testing it in real clinical scenarios. "Many of the results we've returned have had high clinical utility," she said, including for cardiac, neurological, and renal disease patients.
Genome reanalysis offers hope that clinicians can provide even more answers to patients with rare diseases for whom whole-genome or -exome tests did not initially yield a molecular diagnosis or other actionable information.
Generally, diagnostic yields of genome-based testing for various conditions range from about 30 to 50 percent. Stark said that in one study, her team estimated that fewer than 5 percent of patients had any form of data reanalysis, even though literature suggests that the practice can increase diagnostic yield by 10 percent.
"Genome reanalysis" is a bucket term with no universal definition and can mean different things in different places. Sometimes it may mean starting from scratch with the analysis, using only the raw genomic data from previous tests, while other times it could mean asking for reinterpretation of variants of unknown significance (VUS), which may have been upgraded to pathogenic or likely pathogenic, or downgraded to benign.
What started as a process limited to research studies has edged into the clinical domain as studies have shown reanalysis can increase diagnostic yield. "We need to be thinking about how it is integrated into clinical lab practice," Stark said.
Ambry Genetics offers exome reanalysis under its "Patient for Life" program, according to Chief Medical Officer Elizabeth Chao. Internal data show that about 5 percent of initially negative tests will get a new result, with an average wait of about two years. "Because they're new [findings], they're not likely to immediately have a targeted therapy, but they allow for access to clinical trials or other measures of clinical utility," she said.
"We're all acutely aware we should be doing this, but it rarely happens in practice," Stark said. Doing it manually — by having a genetic counselor review the literature on a case-by-case basis — is simply not feasible, with workforce capacity being the biggest barrier. "We're barely keeping up with volume of primary tests, let alone reanalyzing data from patients who've already had testing," she said.
The current crop of programs are not the first attempt to automate genome reanalysis. Stanford University's Gill Bejerano, for example, introduced his lab's Automatic Mendelian Literature Evaluation (Amelie) in 2019. However, "like other tools from my lab, Amelie appears to have been four to five years ahead of its time," Bejerano said in an email. In a 2021 MedRxiv preprint, his lab suggested that Amelie was very efficient, compared to manual reanalysis, being able to make about 80 percent of the same diagnoses while sending no more than one alert per patient per year.
"[It's] great to see interest finally building up for such approaches," Bejerano said.
Focus on genes
According to Chao, new associations between genes and diseases drive more than 75 percent of genome updates. "We do have a program for variant reclassification, but in the context of exome or genome testing, the vast majority of updated reports are due to new gene-disease relationships," she said. "It's much more powerful to look at those."
For Heidi Rehm, codirector of the Broad's Center for Mendelian Genomics, novel gene-disease discoveries and newly classified variants are the driving force for why her lab has installed what it calls the automated interpretation pipeline (AIP). "We normally manually relook at all of our samples every year to see if there are new discoveries to be made," Rehm said. "What we've begun doing with AIP is regularly running the pipeline on our whole dataset." Any new annotations are then loaded into Seqr, a variant search software developed at the Broad Institute. Investigators can see all the new AIP hits across the dataset in batch; alternatively, when reviewing a single case, they can see any AIP updates for that case.
In addition to their research, her lab is beginning to work with clinicians to have their patients’ negative exome or genome tests brought into the research pipeline in a higher-throughput way.
"You don’t have to send a hard drive, one case a time," Rehm said. "They will just give a list of sample IDs for consented patients, and we bring the data on their behalf to a dedicated workspace for their own use on AnVil," the National Human Genome Research Institute's Analysis, Visualization, and Informatics Lab space cloud computing platform. "There's some effort right now to dockerize this pipeline," she said, adding that her lab is talking to Boston Children's Hospital to help them run AIP on their own data in their own installation of Seqr.
Reanalysis Down Under
For Stark, automation of genome reanalysis has two dimensions: incorporating new gene-disease relationships from updated databases and reevaluating a patient's genome in light of that information.
Her project's software does both, tapping into several key sources of new data: ClinVar for new variant-disease relationships, gnomAD for population data on variant frequencies, and PanelApp Australia, a nationwide database for gene-disease relationships. That corpus of data on genes and variants gets updated monthly, upon which the clinical database of patient genomes is automatically reanalyzed. "It now has preliminary data we've started returning to clinicians," Stark said.
Population genomics resources such as gnomAD are increasing the rewards of genome reanalysis by any method, according to Victoria Parikh, a cardiologist at Stanford University. "Population genomics datasets — particularly ancestry-specific population resources — are expanding exponentially, and that has truly informed our understanding of pathogenicity," she said. ClinVar is essentially crowd-sourced, she noted, so "that's hard to use as a gold standard."
Stark responded that how such resources are incorporated into automated reanalysis pipelines "is indeed challenging and requires balancing between the quality of the evidence versus sensitivity and timeliness."
Payment challenges
Stark's project has funding to provide more than 10,000 rare disease patients and their family members with automated genome reanalysis. As part of the study, her team interviewed clinical lab professionals in Australia to find out more about their attitudes toward automation. In a paper published in the European Journal of Human Genetics in May, Stark and her coauthors highlighted concerns about funding, namely that the country would need to come up with a new model for paying for automated reanalysis. Since 2020, a government grant has covered genome reanalysis for patients under the age of 15; before that, it was covered by a mix of clinical services, state government funding, research projects, or self-pay, they wrote.
"This needs to be part of background infrastructure and funded across thousands of patients," Stark said. "Coming up with a funding model will be very challenging." A key part of future research will be to evaluate the economics of automated reanalysis and compare it against the current model, which is "really expensive," she said. "We will be keen to work through the data. That is what's going to inform decisions."
The payment challenge also exists in the US. "One of the biggest barriers is that right now, payors do not reimburse for analysis time," Ambry's Chao said. "We can't bill or get reimbursement for doing this reanalysis, which, in our experience, patients are tremendously grateful for."
"It's rewarding to be able to go back and say 'we didn’t stop searching, we went back and were able to find an answer,'" she said. "A negative result is not the end of the story. This data has value throughout the patient’s lifetime."