Genome.one sequences and interprets genomes to provide diagnoses to people affected by genetic diseases or who have a predisposition to specific genetic conditions. What separates them from the likes of 23andMe and Ancestry.com's services is that Genome.one's goal is to provide information that is meaningful in a clinical sense. At the recent AWS Summit held in Sydney, I spoke with Associate Professor Marcel Dinger, from the Garvan Institute of Medical Research and CEO of Genome.one, about the evolution of this field and how cloud computing has been an enabler as the company navigates the risks of using cloud systems for such personal data.
"Genome.one's entire service is clinically accredited. We have ISO accreditation for medical testing. That means the information we collect and the data is secured in the same way as any other medical data," said Dinger.
This includes an informed consent system for patients so they are aware of how their data is processed and whether they wish their data to be used for medical research. This is different to 23andMe and Ancestry.com, which Dinger says have very broad rights about how the data collect is used. He also said those services only look at about 0.1 per cent of the genome whereas the clinical nature of Genome.one's service sequences everything - about six billion characters of the genome.
When genetic information is sent to the AWS cloud, it is completely deidentified. Data is anonymised before being sent to AWS for processing and then restored once it's back on Genome.one's local systems. Part of the company's accreditation process required that the cloud systems were as secure, or in Dinger's view more secure, than an in-house environment.
Many of the services used by Genome are supported on local AWS regions but some require US regions according to Dinger. They also partner with a company called DNA Nexus, in the US, who use AWS services in the US.
One of the interesting projects Genome.one recently completed was an analysis of the genomes of 3000 healthy people aged over 75 who were free of cardiovascular and neurological diseases, and cancer. That data is used a reference set of data for comparing other genomes in future analyses.
This is the challenge of clinical genetic testing. In order to detect an anomalous gene or some other genotype that is likely to result in the expression of a harmful condition, clinicians need to compare what they seen with reference data. Creating pools of reference data is very resource intensive and requires massive pools of storage and compute power. But that data allows Genome.one to screen for both the disposition for a condition to be revealed or for a specific condition that has already been revealed.
"For a big compute job like that, we were using 4000 servers in parallel on AWS. That just wouldn't be possible in an institutional computer infrastructure," said Dinger.
The information that comes back from a clinical genetic test needs to be reviewed carefully. For example, specific mutations in the BRCA 1 and BRCA 2 genes point to a high likelihood of a woman contracting breast cancer before the age of 60. That information can be used to take some sort of action to protect the patient before the gene is expressed.
Dinger told me the focus of the clinal screening was not to detect every single possible genetic issue. There are many genetic combinations and mutations that can lead to the expression of potentially harmful conditions. But the likelihood of those genes being expressed may be exceedingly low. The clinical focus, he said, was on "highly penetrant" genes where the chance of expression exceeded 50 per cent. That's in contrast to services like 23andMe, which he said provide information such as "you have a 5 per cent greater chance of contracting a condition that the rest of the population" rather than the chance of the gene being expressed at all.
"We don't report on anything that would result in you shrugging your shoulders. We report on things that are clinically actionable. That might be lifestyle changes, such as cardiovascular decisions, where knowledge could prevent something from happening, or medical intervention such as taking statins" said Dinger.
The process conducted by Genome.one starts with a sample of either skin cells or blood. That is sequenced using a machine and the data is sent to systems that operate on AWS. The data is crunched against the reference data and then reviewed by a trained clinician. This is like the pathology service or radiographer that collects data from you, and then conducts and analysis to provide advice to you.
The ability to use technology is driving down the cost of sequencing a genome. At the moment, it costs a couple of thousand dollars to sequence an entire genome according to Dinger. But that is coming down.
"Generating the data is becoming rapidly democratised," he said. "We can see that it will cost next to nothing in the next five to ten years".
Dinger did mention some opposition to predisposition testing. In cases where people have no family history of a specific condition, there are concerns that predisposition testing could lead to over-diagnosis that results on further testing and anxiety for the patient. The goal, he said, was about triaging patients to take pressure off the health care system through early intervention.
The future is what Dinger called "point of care genomics". When we reach that point, the genetic testing can be carried out in a doctor's office with the results delivered back during the appointment enabling a doctor to diagnose and treat conditions far faster than today at a much lower cost. We could reach a point where routine tests such as blood pressure and respiration are complemented with genetic screening to determine your likelihood of something like cardiovascular disease before it becomes acute.
But the amount of computing power to do that will rely on cloud-based systems that can pool massive amounts of power cheaply - something that Genome.one is developing now.