Biotech startup Jungla is the first HPE customer to use the new Memory-Driven Computing Sandbox, to accelerate processing of individualized genomic testing data by more than 200 times compared with traditional hardware, taking the time needed to get results from around 250 days to as little as a day.
MADRID – At HPE Discover here last week, Hewlett Packard Enterprise publicized some of the key applications of their Memory-Driven computing technology. The highlight was a new collaboration with San Francisco-based biotech startup Jungla, which uses HPE’s new Memory-Driven Computing Sandbox to process Jungla’s data sets at extremely high levels of scalability, to let them process vast arrays of clinical genetic and genomic testing data in a timeframe that makes the results useful.
Jungla is the first HPE customer to use the Memory-Driven Computing Sandbox, to accelerate the processing capabilities of their Molecular Evidence Platform.
“We build and deliver computational and accelerator models that drive precision medicine,” said Carlos Araya, Jungla’s CEO. “We have these cellular computation models that are able to answer questions for each patient in terms of genetic tests. There has been a millionfold increase in efficiency since genome sequencing was discovered, but the ability to use that data has not kept up.”
While the costs of genome sequencing has benefited from that efficiency increase – decoding the first genome cost $2.7 billion, while today genome sequencing costs under $1,000 – the bottleneck has been the ability to process it.
“We drive quantitative modelling for all these mutations which have not been seen before,” Araya said. “To do that, we have to be able to crunch through millions of mutations of thousands of genes. Our challenge is to take mutations from any given patient and frame it against everything we know about mutations in that gene and in that disease.”
While the genome sequence of an individual patient – describing hundreds of thousands of variants from thousands of genes – only uses around five GB of information, the volume of the data sets from Jungla’s workflows can require processing more than 40 TB of information for a single gene. The volume of the data sets mean that doing their quantitative modelling on traditional hardware would take so long to process, that the solution wasn’t very medically useful, because it took eight months to do the processing.
“Our Memory-Driven Computing Sandbox pulls in 48 TB of memory, from the combination of our Superdome and the fabric that we acquired with SGI,” said Kirk Bresniker, Fellow and VP. Chief Architect, HP Labs. “This compares to the 1-2 TB of memory that Carlos’ team was used to, next to a GPU. It lets them take these experimental results and stand them up.”
“We have been able to achieve increases of over 200-fold in the time that it takes to do those comparisons,” Araya stated. “We can bring very detailed state of the art computational techniques into clinical environments, where it doesn’t take 250 days to go through the data. Taking 250 days to do it isn’t very useful. This brings the most sophisticated computational model that can be done to provide individualized on-demand care for patients. The other place where this is really significant is adding a mechanistic understanding that helps transition clinical observations into therapeutics.”
“Using the Sandbox enables a transition from a lab process to an industrialized process,” Bresniker added.
In the second part of this announcement, HPE highlighted other work it is doing with researchers in life sciences and clinical medicine from the Living Heart Project and the German Center for Neurodegenerative Diseases [DZNE]. It also followed up an announcement from late September, when HPE partnered with the World Economic Forum to announce Tech Impact 2030, designed to spur collaboration between the technology industry, government, and research organizations to create major change by 2030. HPE and the Forum announced an initial collaboration with Purdue University’s College of Agriculture to facilitate digital and precision agriculture possible and greatly ramp up food products.They have now launched their second Tech Impact 2030 challenge – to enable real-time, personalized medical care for patients using Memory-Driven Computing.
Another collaborator is Dassault Systèmes and their Living Heart Project.
“Our premise was to take on the number one cause of death, heart disease, and see if we could replicate the human heart on a computer and apply powerful tools to assess it,” said Steven Levine, Senior Director, Dassault Systèmes Health and Life Sciences, and Executive Director, Living Heart Project. “We set out to bring scientists, doctors and other specialists in, and launched in 2014. By 2015, we had built a fully-functioning human heart which was available for sale, Before, only elite scientists in national computing centre could use something like this. Now anyone with a workstation in a hospital can.”
Levine said that the ongoing research is doing the same thing for the other parts of the body.
“The goal of everything we are doing is to take generic medicine and make it personal,” he said. “We are now expanding to looking at the brain, eyes and the rest of the body. What we haven’t done is efficiently translate it for every individual – like has been done with the human genome. At one point, it wasn’t cost effective to break that down to the level of individuals. Now it is. We just need to get good at this.”
“HPE and the DZNE have been partnered for three years, and we have now taken it to the next level,” said Joachim Schultze, Director, PRECISE Platform for Single Cell Genomics and Epigenomics, at DZNE and the University of Bonn. Schultze views memory-driven computing as a key to the emerging sophisticated compute infrastructures necessary for precision medicine that can make real breakthroughs.
“By 2030 we would like to have stopped Alzheimers from being an incurable disease, and turn it into a chronic disease that can be managed,” he stated.