Deep learning is making technological progress at breakneck speed.
One day reinforcement learning is proclaimed the Holy Grail of general AI; the next, Geoff Hinton is publishing a new paper about capsule networks. AlphaGo can probably beat every Go player in the history of Go and might soon be playing baseball.
But there’s more to AI than big names and papers on novel algorithms. This week, Humans of AI had the chance to talk to Pete Kane, the founder of SVAI, an AI Life Sciences community in San Francisco. With his organization, Pete is working to improve the industry from a different angle: forming a community of researchers and entrepreneurs and bringing them together around the most complex medical problems of our time.
From Minneapolis to Silicon Valley
Pete started his career as an entrepreneur by founding several startups in healthcare technology while studying Chinese literature in college. Although most of these companies didn’t turn out to be very successful, they provided Pete with the experience he needed to create Healthcare.mn, a community for innovation and startups in the Minnesota healthcare technology industry. “It became pretty popular,” Pete remembered, “and a center of healthcare innovation in Minnesota.” It was here that he laid the foundation for his later work in community building.
Though he was happy about the success, after almost 3 years at Healthcare.mn Pete was eager for new experiences: “I needed to do something new, so I moved out to Silicon Valley, kind of on a whim.” Without a real idea what to do in the Valley, Pete began to investigate the local tech communities and compared them to what he knew from back home. He soon realized that, while AI is a hugely popular topic in Silicon Valley, it seemed that the social events surrounding the field were lacking a sense of community.
“At the time, there were a lot of new AI events and meetups popping up. There were a few good ones, but most of them were underwhelming,” Pete said. “People just piled in and after the speakers were done speaking they piled out. There wasn’t a lot of sense of people getting to know each other.” From his community days in Minnesota, Pete knew there was potential to create something better, a meeting format that could actually build a connection between participants.
Creating a New Kind of Community
With these thoughts in mind, Pete attended a talk that Richard Socher gave at Baidu. Socher was the founder and CEO of MetaMind, an AI startup that was later acquired by SalesForce. During the talk, Socher mentioned their partnership with a healthcare technology firm in Minnesota. The Minnesota company was providing radiology images for MetaMind to train their deep learning models.
It was a key moment for Pete: “I thought that was pretty cool that some of the top Silicon Valley folks in AI were looking to the Minnesota healthcare scene for data. That’s when I felt like I could really add something to the AI field here.”
In 2016, Pete founded SVAI, a community for AI researchers and engineers. He wanted the group to address the shortcomings he had observed in other AI gatherings. Instead of inviting the big names in the industry to talk, SVAI focuses on younger, up-and-coming researchers and leaders in the field.
“We feature Ph.Ds., younger researchers, and folks earlier in their careers,” Pete explained. “They’re a lot more accessible, you can go grab a coffee together … there’s tons more potential for collaboration. We’re growing as a community with the people we feature at SVAI.”
In the beginning, the group organized general AI talks. “We did a few self-driving car events, as well as an event on AlphaGo right after the big game,” Pete said. But soon, he decided to direct the group’s focus towards the topic he was most passionate about: healthcare and life sciences. It was a good fit for the group and an unexplored niche: “There was no other group that was giving this intersection a lot of attention.”
Hacking on Real Genome Data
In 2017, SVAI organized their first hackathon on AI and genomics. Supported by Google, NVIDIA, NIH, and several other companies, the participants spent a weekend exploring genetic data on neurofibromatosis type 2 (NF2) and looking for new insights to treat the rare genetic disease.
The dataset used for the hackathon consisted of real data from an ongoing medical case. Onno Faber, an entrepreneur diagnosed with NF2, used his own funds to sequence his genome and provided the dataset for the hackathon.
The NF2 hackathon was well received, and Pete is now planning another hackathon in May. The second event will focus on Papillary Renal Cell Carcinoma Type 1, a rare kidney cancer. SVAI will again provide real genomic data: “We’re partnering with a patient initiative and working on getting a patient’s tumor sequenced for the event.”
Creating open and free biomedical datasets has become a focus of SVAI. “The number-one complaint that we hear over and over again is that so much data is locked up in institutions or academia,” Pete said. In part, the reason for the scarcity of available genomic datasets is the cost of creating them.
“For the first research hackathon, the patient spent about USD $45,000 to get his tumor, his blood, and his brother’s blood sequenced,” Pete explained. Since then, the prices have dropped, but it remains expensive to create large sets of hundreds of samples.
SVAI is Creating Open Access Datasets
SVAI is currently working on “finding unique ways to fund and create novel, larger datasets,” Pete said. He explained why it is difficult to use existing genomic data: “For any datasets created from a human, the person has consented to a clinical trial to collect this data.
None of those consents allow for the data to be free and open to the public.” Thus, instead of using existing data, SVAI is planning to create new datasets specifically for this purpose: “[These] datasets will be 100% open access and free to the public perpetually.” One of the datasets Pete wants to create is an HIV Genomics dataset to support viral genomics research.
“What drives me,” Pete said, “is creating new opportunities for extremely bright AI engineers and researchers.” He is optimistic about the future: “There’s enough [to do] for an entire lifetime of research at the intersection of computation and life sciences.” And SVAI, Pete hopes, can play a part in this by bringing people together and providing high-quality biomedicine data.