Finish What We Started: Dietrich Stephan’s Silicon Valley Biosystems Opens Genome Interpretation Floodgates

June 12, 2012

By Kevin Davies  

June 12, 2012 | SAN FRANCISCO -- The co-founder and former chief science officer of Navigenics, Dietrich Stephan, says he aims to “finish what we started” with the commercial launch of a new Silicon Valley genome interpretation company, Silicon Valley Biosystems (SVBio).  

Stephan briefed Bio-IT World exclusively on the eve of his official announcement at The Clinical Genome Conference (TCGC), a new scientific meeting devoted to advances in clinical diagnostic sequencing, organized by Bio-IT World and Cambridge Healthtech Institute, which opened today in San Francisco. 

The SVBio announcement signals a new phase in the increasingly competitive battleground for medical genome interpretation software development. Among such companies presenting at TCGC this week are Knome, which just announced knomeCLINIC; Cypher Genomics, a start-up co-founded by Scripps physician Eric Topol; Personalis, a well-funded Stanford start-up; Genomatix and many others including Omicia (a Bio-IT World Best Practices winner), Ingenuity, and Station X. 

“Despite Navigenics not being the monster company we wanted it to be (due to actuarial data being slower to come in chronic disease diagnostics), we built the industry and wanted to hit the refresh button and do the necessary inventing around rare variants now that the time is right,” says Stephan. (Navigenics dropped the direct-to-consumer approach for its personal genomics services years ago. Stephan says he received the company’s permission to launch his new company, though he has since stepped down as Navigenics CSO.) 

The name SVBio was originally considered just a placeholder but Stephan, who is the sole founder, president and CEO, has grown to like the name. The company is a software company generating and transmitting CLIA-grade reports on patient genomes to physicians, who will also have the option of using SVBio’s clinical sequencing labs. SVBio has established four centers conducting clinical exome sequencing work, based in Mountain View, CA; Rockville, MD; Research Triangle Park, NC; and South Korea to ensure it can service the global marketplace. 

“This is about creating a large and sustainable business,” says Stephan. “The strategy for the product road map and the testing types – gene panel, exome, WGS, etc. is something we’re thinking carefully about. We’re starting with exomes – we’ll have 3-4 different offerings. The end game is to be a software company and disconnect from the lab completely.”
Stephan itemizes what the final clinical report might look like: “Page 1 has answer on it. Page 2 has variants in genes that may be modifiers or confounding of disease. Page 3 has the hard-core quality control metrics. And Page 4 has the disclaimer and signature.”
 

Ignition SequenceAfter Stephan’s ambitious plans for the Ignite Institute for Individualized Health were scuppered a few years ago, he refocused on his entrepreneurial interests, including an advisory role for the California nanopore sequencing firm Genia Technologies, and crafting the business plan and obtaining the funding for the Children’s Hospital Boston’s Gene Partnership Project. Another start-up was a cancer diagnostics firm called Aueon, which he founded with intellectual property lawyer Vern Norviel, whose wife had developed cancer. Aueon didn’t attract funding, although Stephan says it does own core intellectual property for somatic tumor testing and therapy selection.  

Over the past 12 months, Stephan has been quietly building SVBio. Initial funding comes from Sequoia Capital, which also funded Navigenics – not to mention Apple, Oracle, Google and LinkedIn. “They have a reputation as the top VC firm,” says Stephan. “Their model is to look for transformation trends in the world and try to figure out value around those.”  

Sequoia brings a hard-core business sense to the initiative, he says. “They’ve been watching the sequencing space and recognizing that pattern. There’s something there, but where’s the value? They quickly recognized [genome sequencing] was a commodity business that would be high volume, low margin, and difficult for a new player to build a stand-alone business.” 

Nor was there huge value to be found around DNA alignment/variant calling tools. Instead, Sequoia quickly settled on the clinical market. “It’s a massive market, not like the research market,” says Stephan. “The value continues to lie in moving from a variant to a doctor sitting in front of a patient in the clinic as we anticipated in 2005. The technology has finally become accurate and cost attainable.” 

Stephan says Sequoia aims to build “the trusted clinical brand for a new era of diagnostics.” But how will he become the fulcrum of next-gen clinical diagnostics in an increasingly crowded marketplace? “We’ve put together a SWAT team of ten really smart people, including computational biology and genomics, to build out a platform for generating next-gen data.” Stephan’s chief advisors and colleagues include:
   • Eleazer Eskin, a statistical geneticist/computer scientist at UCLA;
   • Timothy Triche, a molecular pathologist and principal investigator at the Saban Research Institute at Children’s Hospital, Los Angeles;
   • Sivan Bercovicci, a geneticist and machine learning computer scientist; and
   • Eugene Fratkin, a geneticist and machine learning computer scientist
 

Stephan says his team has identified the common and rare variants that cause almost 30 human diseases. “We have a visceral feel for the genome, and understand how to codify that experience computationally,” says Stephan. For example, Triche’s team at OncorMed patented the first four sequence variants associated with breast cancer predisposition, eventually licensing the IP to Myriad. “This work changed the world of diagnostics,” says Stephan.  

But SVBio is not just about new software to sift through the genome. The company has established a full high-throughput clinical sequencing lab, performing exome sequencing using the Illumina platform, Agilent target selection, and full QC/QA at high coverage. “If the sample doesn’t meet [our standards], it goes back,” says Stephan. 

Stephan says the output is high: the group can pump out samples at a rate of 300,000 exomes/year already. The decision to bring on-line sequencing labs is primarily to ensure clinical grade quality, as well as to better understand the torrent of data, the nature of systematic sequencing errors, and to learn about building the critical data storage, local and cloud compute, to manage the data. Amazon Web Services has been great, says Stephan, but his colleagues are still sending hard drives to Amazon. “We’ve got a couple of hundred nodes locally; we just need the cloud for surge and special projects,” he says. 

All of the samples sequenced so far are from undiagnosed patients with putative monogenic disorders. Deals are in the process of being inked with a number of prominent clinical centers. SVBio plans eventually to potentially phase out the lab services as the technology stabilizes into accurate long-read sequences and focus on “a clinical-grade software package.” Rather than producing software that runs on top of the sequencing instruments, Stephan says, “We decided we’d go to where the people who do the interpretation and testing today are as our point of contact.” 

The first product released is built around monogenic disease applications. “Taking the end result of an exome and annotation exercise – you’re usually left with 200-300 [candidate variants]. How do you take in prior information about the phenotype of that patient, eliminate the false positives, and spit out the causal variant at the back end?”  

“Our solution is highly differentiated from what is on the market today. For clinical work a 30% false negative rate and huge false positive rate, which is the norm using filtering-based products, is totally unacceptable,” says Stephan. Our solution offers the ability to move from sample to variant in a turnkey fashion. 

Hit RateThe current lab turnaround time is four weeks, but Stephan aims to halve that in the next 12 months. The alignment and variant calling takes several hours, while the actual proprietary interpretation analysis takes minutes to produce the final result.  

SVBio’s partners have evaluated the accuracy of the more than 200 clinical samples analyzed so far. In one test utilizing true clinical samples where mutation status was known, “the genome to mutation (G2M) product’s sensitivity and specificity exceeded 99%,” says Stephan. 

Noting that there are fundamental problems in detecting large copy number deletions and trinucleotide repeats using short-read sequencers and current aligners, SVBio has begun testing a proprietary copy number variant module for exomes. The company will be looking at cancer predisposition but not somatic tumors, says Stephan. Over the coming 3-5 years, SVBio plans to expand into pharmacogenomics, complex risk assessment and/or host-pathogen interactions. 

Stephan and Sequoia doubt that the big boxmakers are capable of building first-class clinical software. “It’s a totally different skill set,” claims Stephan. He notes, for example, that rather than build in-house OEM software, Illumina has created a clearinghouse for third party solutions.  

But Stephan will still have to prove that SVBio can rise above the challenges posed by a wave of emerging genome interpretation companies including some, such as Knome, Omicia and Genomatix, which have been around for a number of years.