This is part two of a series. The previous story is here.
O n a Friday summer evening in San Francisco, 150 smart young people flooded one floor of a South-of-Market office building and filled the space with their excited voices. Onno Faber, a Dutch tech entrepreneur with tousled blond hair, flitted among them, leaning in close with his right ear to hear each person above the hubbub. He was the reason they all had come.
“I don’t know anybody,” Onno said of the guests, who had traveled from universities and pharmaceutical companies across the country. Then he laughed and added: “But they will know me.”
The crowd had come to spend the weekend crunching the data of Onno’s DNA. They’d be searching for genetic clues in hopes of curing his neurofibromatosis type 2, or NF2 — a rare disorder that had already stolen the hearing in his left ear and caused the tumors now growing slowly in his right hearing nerve and spine. Eventually they could wipe out the rest of his hearing, his sight, and his ability to walk.
This hackathon was the first public event in Onno’s quest to use the power of modern computing to do his own medical research. And it demonstrated perhaps the biggest idea in medicine today, the blending of computer technology and biology to crack the codes of disease. As genome sequencing has become more affordable and accessible, computers have grown powerful enough to analyze the staggering amount of data that such sequencing reveals. Understanding the workings of certain diseases is becoming a very complicated pattern-recognition problem.
Onno’s hackathon was unique in that it was organized and attended by the patient whose data was on display. But it was one of many going on these days. In fact, the application of advanced computing to big genetic data is so intriguing that the National Institutes of Health has an expert, Ben Busby, who devotes half his time just to traveling the country running genomics hackathons. Other organizations run such events in Europe and Japan.
Busby welcomed the volunteers at Onno’s event with the booming voice and exuberance of a carnival barker. “We are going to try to make a difference in the world, and that’s awesome,” he told them.
Onno, who is 35, started his first Internet company while in high school and now lives and works among fellow technologists in San Francisco. For him, genome hacking is a natural way to try to solve his disease, which has been relatively neglected by the pharmaceutical industry because it’s rare. As the hackers in San Francisco filled out their name tags, he held up a thumb-sized hard drive containing his genome and remarked, “People were saying, ‘You’re sitting on your data. Why?’”
The hackathon had been organized by technologists, led by Onno and a nonprofit called Silicon Valley Artificial Intelligence that focuses on applying AI to life sciences. Google contributed a physical space (its Launchpad location) and digital space ($150,000 in cloud computing power). But the event also caught the attention of serious biologists.
While the hackers munched on free pizza, Marco Giovannini, director of the Neural Tumor Research Laboratory at UCLA and a prominent NF2 researcher, addressed the crowd. They included graduate and undergraduate students in computation and biology, plus professional bioinformaticians, artificial intelligence engineers, and young workers from the pharmaceutical industry.
Giovannini showed them a research paper proposing that, in addition to the gene mutation known to lead to NF2, there could be something else — more mutations — involved. Then Giovannini, who has spent more than 20 years studying NF2, turned to the room full of relative newcomers and put to them the question to which he’d like an answer: What is that something else?
And so their treasure hunt began.
The hackers had three full genomes to work with: one from Onno’s tumor, which contained the known NF2 mutation. One from Onno’s blood, which is free of that mutation. And one from Onno’s brother, who does not have NF2, for comparison.
“You have millions of mutations, right? So you have to find the ones that you think are responsible for the tumors,” Onno explained.
The three data files were so large (nearly a terabyte in total) that Onno couldn’t find a workable way to send them to participants to download before the hackathon, so ultimately he had the participants access them in a Google cloud. Remembering the days of his first Internet company in the 1990s, when it was hard to transfer a simple image with a dial-up modem, he said, “I feel that way today about transferring genomic data.” Biology is still complex enough to dazzle technologists.
Busby, who is genomics outreach coordinator and bioinformatics training lead at the National Center for Biotechnology Information, started running genomics hackathons in 2015 and now does them monthly. Many are cohosted with universities or other research institutions. None has been as well attended as Onno’s, which offered a chance to hack someone’s DNA with the person present. More than 400 people applied for the 150 spots.
The purpose of such events, Busby says, is both education and production: teaching biologists about informatics and informaticians about biology, while also generating algorithms and analytical tools to fuel further work. The fruits of all these efforts are open source, posted online for anyone to use.
One cancer researcher said the experience was “almost a dream come true, because I had no idea machine learning could be applied in such a manner.”
Busby keeps a scientist’s conservative outlook when he’s not giving pep talks to hackers. He’s clear that no single hackathon can be expected to produce a cure for a disease. Even the discovery of a potentially relevant gene mutation is only the first step on a long, expensive road through further research and eventually clinical trials to reach a treatment.
But despite that cautiousness Busby believes in the potential of genome hacking to help us better treat disease. The genetic nature and behavior of tumors can vary from patient to patient within the same disease; it can even vary from tumor to tumor within the same patient. Busby hopes that, by building knowledge, we can eventually differentiate tumors based on their source code and thus treat them more precisely and effectively.
“What we’re talking about really is speeding up the time scales of science,” he says. “Genomics right now is more or less the Wild West. There’s new stuff coming out every single day.” That’s why it’s especially valuable to build technical tools that enable anyone to analyze genomic data faster. “And we’re pretty sure that the biggest user of the things we make in the hackathons are industry companies. We see them scraping our stuff all the time.”
Onno wanted something much more personal: clues to a cure for NF2. He also wanted confirmation that collaborative data crunching could produce something useful.
He got both. Twenty-plus hackathon teams clustered around tables all weekend and worked until the doors closed at 11 p.m. on Friday and Saturday nights. Over dinner on Saturday, one team told Onno they’d used an open-source drug-discovery tool called DeepChem to identify a chemical that could target a protein related to NF2. Another team used machine learning to figure out that genes known to play a role in other cancers are also active in NF2 tumors. That might open up new avenues for treatments.
That team’s leader, Jyotika Varshney, a postdoctoral cancer researcher at the University of California, San Francisco, said the experience was “almost a dream come true, because I had no idea machine learning could be applied in such a manner.”
By Sunday afternoon, however, Onno’s initial excitement had turned to sober determination. Alone again, he leaned his forehead against his hand, furrowed his brow, and gazed into the distance. Following Saturday’s high, he said, “It became very real again. Which was also very cool. I could hardly wait until it was over, because I have so much to do and I want to get back to work.”
Giovannini, the NF2 researcher, said the promising chemical and other findings from the hackathon could prove fruitful — but only if Onno and his collaborators collect more genomic data from other NF2 patients and see whether the patterns hold in other people. That’s Onno’s goal for November, when he and Silicon Valley Artificial Intelligence are planning a second hackathon in New York City. Now he is reaching out to fellow patients from around the world, encouraging them to have their tumor genomes sequenced and share the data.
Busby is considering whether to be there in November, too. “Imagine what that group of people could have done if there were a thousand Onnos worth of data,” he said of the first hackathon. “It would be staggering.”
Neo.life will be following Onno Faber’s story as it unfolds.