Exclusive: Pfizer vet grabs $25M seed round to 'integrate' machine learning, genomics tech in drug R&D
Having been involved in a number of attrition task force efforts during his tenure as European R&D chief at Pfizer, David Roblin tells me there’s a quote that resonates with him, uttered 20 years ago by South African biologist Sydney Brenner when he accepted a Nobel Prize.
“We are drowning in a sea of data but thirsty for knowledge,” is how Roblin remembers it.
But the way those data have been used to date — by drug discovery scientists trying to validate their ideas — hasn’t really improved the rate at which preclinical compounds become marketed drugs, he reckons. The potential he sees for machine learning and new technologies like single-cell transcriptomics to change that is the reason why, after eight years as chief of scientific translation at London’s Francis Crick Institute, he’s moving down the road to Relation Therapeutics, which has just closed a $25 million seed round.
“We’re not going to sit here and tell you that machine learning is the single answer and everything will be designed in the machine and that machine will send a file to the FDA,” he said. “The key question for a company like ours is, where do you best deploy machine learning that affords maximum impact? Where do you use transcriptomics? And where do you see the linkage?”
Roblin, the CEO, is joined in the C-suite by Lindsay Edwards, CTO and president of platform. The former head of AI for respiratory and immunology at AstraZeneca, Edwards also led one of GSK’s first data science groups before that.
Coming out of stealth at a time the idea of deploying AI/ML in drug R&D has already gone through a couple cycles of hype and bust, Relation’s approach will be grounded in active learning, where machine learning systems will go through the data generated in its wet and in silico labs, and then tell the scientists what new experiments to run for it to come up with new insight.
Specifically, Edwards noted, the company will be using a framework called active-graph machine learning, or Metagraph, using the concept of graphs as a backbone so as to map out the complex relationships between genes, proteins and drugs.
“What Lindsay is doing, largely, is taking algorithms, ideas that have been deployed elsewhere, particularly in consumer social networking, and turning them to the purpose of gene variant biology, identification,” Roblin said, adding, “So in that sense, I’m quite happy to say like, a lot of that stuff is not absolutely breakthrough. What we’re going to be doing is creating a platform of integration which is second to none, essentially.”
Based in the Knowledge Quarter in London — surrounded by UCL, King’s and Francis Crick as well as several hospitals — Relation is looking to crunch genomic data on the individual cell level with samples provided by neighboring biobanks.
As its first big project, the team is developing a cell atlas of the bone, an area where significant gaps in biological understanding still exist, according to Roblin. It is important, in his words, to fully grasp the translational route and ensure the cell models used in Relation’s labs are “as close to recapitulation of disease as possible.”
“Making sure you work on the right things is a really critical problem,” Edwards said.
Relation will also get help from Nvidia, which is giving it access to the NVIDIA Cambridge-1 GPU supercomputer, which helps opens up new possibilities for looking at large stretches of DNA.
“Everybody runs into a compute limitation, and that basically sets the width of the amount of DNA that you can look at in a single go,” he said. “So access to Cambridge lab, we expect will give us some kind of breakthrough receptive fields with our models.”
With “really good R&D people” at the center of it all, Roblin said Relation will aim to grow from 15 to 40 staffers with the new cash. The company has also put together a hefty scientific advisory board, including Michael Bronstein (Oxford professor and head of graph AI at Twitter), as well as Alex Shalek and Caroline Uhler of the Broad Institute.
DCVC and Magnetic Ventures co-led the financing, with participation from Khosla Ventures, OMERS Ventures, and firstminute Capital, plus a number of individual investors.