Cypher Releases Validation Data on CNVs
By Allison Proffitt
April 10, 2015 | Cypher Genomics released data last month showing that the company’s Mantis software provided “highly equivalent interpretations” to a panel of genetic counselors when evaluating whether 91 copy number variations (CNVs) detected in non-invasive prenatal testing were pathogenic.
The new data comes just a few months after Cypher announced a development agreement with Sequenom. The prenatal testing company is using Mantis to advance analysis of clinically-relevant fetal sub-chromosomal variants detected in maternal blood.
In January, Adam Simpson, president and COO of Cypher, said, “We have been working closely with Sequenom to validate our automated Mantis technology to classify clinically-relevant sub-chromosomal genomic structural variations, such as copy-number variations, in circulating fetal DNA from maternal blood samples, which may have applications in the development of a new, more comprehensive NIPT offering." The data released in March is the result of that validation.
Cypher arrived on the genomics interpretation scene in January 2013 with the launch of their beta access program for clinical genome interpretation. Bio-IT World spoke with CEO Ashley Van Zeeland then, but in the years since the company has been evolving. “We’ve spent quite a bit of time on continued development and validation of a lot of the automated interpretation techniques,” she said.
Today, Cypher is, “focused on the creation of differentiated diagnostic tests by clinical laboratories and the creation of novel genomic-based biomarkers by pharmaceutical companies,” Simpson explained.
The Sequenom agreement falls on the clinical side; a July 2014 deal with Illumina promotes Cypher’s biomarker discovery service in conjunction with the NextBio platform.
Simpson believes that the breadth of the diagnostic market is expanding. Sequenom plans to use Cypher’s software to offer a next generation non-invasive prenatal test covering more of the genome, and Simpson see the same expansion in oncology, where current tests focus on a few hundred genes, but that will soon increase.
The expanding offerings will put even more stress on genomic counselors and clinical geneticists to interpret findings.
“That’s really where Cypher comes in. We’ve been working to validate our automated interpretation that would be used by the clinical laboratory [that is] at least as good if not better than human beings.”
The data presented last month at the American College of Medical Geneticists meeting, shows that Cypher’s software can do just that, Simpson says.
“Mantis is really an automated genome interpretation system to help identify clinically-relevant genetic variants from sequencing data, whether you start with whole genome sequencing data or exomes or smaller panels,” explained Van Zeeland.
Cypher compared Mantis’ interpretations with those of a panel of human experts.
The study looked only at CNVs, which are often not detected until after birth. NIPT has thus far tested for whole chromosome errors—trisomies and aneuploidies. But genetic variations that occur within a chromosome, such as copy number variation, account for 70% of live births with chromosomal abnormalities, Cypher believes. These intra-chromosomal variations are more difficult to classify as disease causing, because similar diseases may be caused by a variety of different mutations.
For this study, 91 detected CNVs were manually curated by working groups of genetic counselors and clinical lab directors. The detected CNVs comprised both duplications and deletions, spanning a range of ~1-77 megabases in size. The initial expert review resulted in 59 of 91 CNVs classified as pathogenic or potentially pathogenic. The remaining 32 were considered variants of unknown significance (VUS).
The results weren’t simply variant classifications. “You can classify variants according to whether they have been reported as pathogenic, VUS or known benign according to various sources, but that is not sufficient for a full clinical interpretation and there are multiple additional decision points along the way,” Van Zeeland explained. “Given the false positive entries in many databases, as well as occasionally weak evidence in the primary literature, even variants classified as “reported pathogenic” mutations are often subsequently ruled out in the final interpretation. Further, variants of low quality, or those that are likely to be sequencing artifacts also must be interpreted carefully and we have computational approaches to the full interpretation process that help our clinical lab partners as described.”
A comparison of the expert’s opinions to Mantis’ automated interpretations showed a high concordance, suggesting Mantis could provide a scalable and automated solution to reporting clinically relevant CNVs. Mantis also was able to characterize more of the CNVs, reducing the number of VUS and highlighting the potential for increased diagnostic information in NIPT via automated classification.
The findings, “prove that we can save the clinical laboratory substantial time to greatly increase their throughput,” Simpson said. “Our experience with the leading clinical laboratories is that just because they interpreted something by hand one time, they will be rechecking that decision periodically. The space is evolving so quickly, there’s so much additional information coming out, there might be new information tomorrow that is highly relevant to the diagnosis. It continues to be a very time consuming process going forward... We’ve taken a very similar approach, except that we’ve automated that.”
Simpson and Van Zeeland have no intention of replacing genetic counselors, but rather saving them time to focus on the findings that most need their attention. For example, Simpson said, it’s well known that there are false positive or inconclusive results in the literature. “We can help identify the cases where medical geneticists might want to spend a lot more time researching a particular variant of interest based on disparate literature.”
Cypher doesn’t return a report with simple scores for each variant, Van Zeeland explained, but also provides all of the evidence that drives that classification, helping users prioritize their time. For example, “This [variant] has been reported pathogenic, but based on other experience we have or other models we have in our system, I would go double check this. I don’t believe it’s truly pathogenic.’ In the absence of that flag, we’re able to have this very high agreement with how they would otherwise interpret it.”
One of Cypher’s internal models is the data from the Scripps Wellderly Cohort, whole genome sequencing data from healthy adults of advanced age. Cypher uses those data to flag potential false positive findings.
“It highlights the fact that there’s a lot of noise in this space,” Van Zeeland said. “In the absence of standardized approaches or the absence of other more sophisticated techniques to clean up that information, a lot of time can be spent chasing down these very expensive false positive findings.”
The errors are primarily because evidence was mis-curated or never corrected, she said. “It’s a problem with reference databases.”
The platform validated on these 91 cases is the same that underpins the rest Cypher’s offerings, Van Zeeland said. “Somatic interpretation capability, all the techniques that go into that, the biology techniques the big data interpretation techniques are derived from the same validation platform that we presented at ACMG with the copy number variant interpretation as well as our experience in germline interpretation that we’ve been validating for a number of years.”
The validation data comes just as Cypher is poised to break into oncology, Simpson said. “Our goal is to support leading clinical labs in genomics-based markets that have shown clinical utility. It’s very important to us to support the oncology market, and we’re very excited to be in the prenatal space,” he said.