PierianDx Plans Software Update to Improve Variant Prioritization, Clinical Relevance Assessment
by Josh Forsythe, on April 2, 2015
NEW YORK (GenomeWeb) – This summer, PierianDx plans to add a so-called PDx score to the next iteration of Clinical Genomicist Workstation (CGW), the company’s proprietary next-generation sequence data analysis platform, that it claims will help clinicians prioritize actionable variants in somatic cancers and constitutional disorders such as cardiomyopathy.
In addition, the company is expanding its underlying knowledgebase to include additional information on clinical trials and treatment guidelines culled and curated from multiple sources, the company said.
The CGW provides applications for processing, analyzing, storing, and reporting on genes and variants identified by clinical next-generation sequencing-based tests. The system also includes a database of curated literature, biological information, computation predictions, population frequency data, clinical research correlations, and disease variants. It was initially developed by researchers in the Genomics and Pathology Services laboratory of Washington University in St. Louis and is still used there. It was licensed by PierianDx in 2014 and is now sold by the company under a software-as-a-service model. The company charges per report generated using its system, roughly 10 percent of the cost of the test or assay performed.
The new PDx score ó as well as the updated information ó will be included in version 3.0 of the company’s software, slated for launch in June this year. It’s a tool that will help medical directors “prioritize or wade through increasing volumes of information [and] assist them in determining the right path to practice medicine,” Ted Briscoe, the company’s CEO, said in a conversation with GenomeWeb this week.
That’s especially important “as we move into more sophisticated [genetic] tests [and] the ability to find the most relevant actionable information from a [growing] knowledgebase becomes a far more challenging task,” he continued. “We’ve put a lot of our focus in not just building the knowledgebase … but also making sure that we have this learning algorithm that can very efficiently and accurately extract what we believe is the most relevant information.”
The PDx score is calculated using a proprietary algorithm that sorts through all of the information contained PierianDx’s internally developed knowledgebase and generates individualized variant scores based on patients’ clinical and genetic data as well as the testing assays used.
The new score moves away from the purely rules-based approach that underlies the current PierianDx system and adopts more of a learning approach, which offers several benefits, Rakesh Nagarajan, PierianDx’s chief biomedical informatics officer, explained to GenomeWeb. With the current system variants are interpreted based on sets of rules applied, for the most part, independently to the different sources of information in the knowledgebase. But the PDx score integrates information in a way that helps the user more quickly understand where those sources agree and disagree with each other and to what extent, he said.
Moreover, the score will also help clinicians better deal with variants of unknown significance, he said. With a pure rules approach, “those variants would simply be variants of unknown significance with no other information from interpretations or publications perhaps,” he said. “Layering in other information that are preclinical as well as computational helps provide the medical director with evidence … that they can then use to evaluate whether they agree or disagree with those preclinical evidences to merit writing something about that variant in the report.”
According to Nagarajan, scores are based on a wide range of information contained in the company’s knowledgebase. That includes “pure computational evidence, all the way through to very well defined clinical pathogenicity and everything in between, [such as] in vitro evidence, animal model evidence, human research evidence, clinical trial evidence, emerging guideline evidence, publication evidence, and established clinical practice.”
The system gives variants relevant to a particular patient’s case a score from zero to one; and variants that have established clinical utility and pathogenicity are accorded higher scores, he said. So for example, an EGFR variant in exon 19 has known targeted therapies in non-small cell lung cancer, so if this variant showed up in the test results of a patient with the non-small cell lung cancer, it would be awarded the highest possible PDx score in this instance, he explained. On the other hand, if that same variant were found in a patient with another type of cancer, where there aren’t established treatment guidelines, the score would be slightly lower, he said.
PierianDx’s database contains information from sources such as ClinVar, COSMIC, TCGA, dbSNP, HGMD, and the 1000 Genomes Project, Nagarajan said. It also contains computational evidence from tools such as Condel, SIFT, and Polyphen, and content provided by members of the company’s partner network, which includes Phoenix Children’s Hospital, the University of Arizona, and ARUP laboratories.
The company can also incorporate information from proprietary databases such as HGMD, if clients have existing licenses to these resources, and use the data in its scoring mechanism.
Meanwhile, PierianDx is also expanding its knowledgebase to include curated information on clinical trials, FDA drug labels, and treatment guidelines from the National Comprehensive Cancer Network and the American Society of Clinical Oncology.
These added datasets will provide more comprehensive information on variants associated with cancer and constitutional disorders than was previously available to clients working with cancer patients, Nagarajan said. To date, within the somatic cancer space, the medical interpretations of the variants identified in patients’ tests came from medical directors participating in the PierianDX partner network.
“That’s very valuable information that can be immediately and directly [used] by other directors should they agree completely with that interpretation,” he said. However, “it was very opportunistic in that a variant that had never been encountered by our director network would not have an interpretation, and those interpretations would have to be written from scratch based on our curated publication.”
The updated database will provide a more comprehensive curation of variants associated with specific cancers matched to relevant guidelines. “So now if an interpretation doesn’t exist, the guideline information exists [and] that information can then be used to draft an interpretation … that can be used for interpretations in the future,” Nagarajan said. “Similarly, if guidelines don’t exist but there are publications that are the first pieces of evidence that come out, that information can be used to write interpretations.” The same principle applies to clinical trial data.
For customers in the constitutional disease arena, the updated system will offer a deeper curation of information in resources such as ClinVar, Nagarajan said. It will also include variant information from the EmVClass database maintained by Emory Genetics Laboratory, the ClinVitae database, as well as content from network partner ARUP labs.