Atomwise's X-37 spinout gets $14.5 million to launch AI discovery efforts
The folks behind Atomwise’s spinout X-37 like to think in cosmological metaphors, and you can think of their AI drug development model as probes sent into space from a central station. That station just got $14.5 million in Series A funding from DCVC Bio, Alpha Intelligence Capital and Hemi Ventures to back those missions.
X-37 uses Atomwise’s AI platform to identify drug targets and – unlike the parent company, which largely sticks to computers – bring those into a wet lab and preclinical testing. In addition to AI professionals, it’s led in by part by drug developers from Velocity Pharmaceutical Development.
Their business structure is noteworthy and becoming increasingly common for biotechs. Each drug development program is housed as its own virtual company. CEO David Collier used the model at Velocity and said it allows X-37 to cut any failed targets without affecting the rest of the company, or licensing out the drug and potentially running into intellectual property issues down the road.
“Twenty years ago, if you funded a biotech, and you had 3 to 4 programs, you ran into the problem that a pharma company wanted one of them but not the rest,” Collier told Endpoints News. “Your options were to license the drug or sell the biotech and give up on everything else. So people have been moving to the LLC structure.”
The name X-37 is a reference to Boeing X-37, a reusable spacecraft that has made five trips into orbit and returned last month after a record 780 days in outer space. On their website, they explain their vision in space meme format:
In that theme, you can imagine their targets as different probes out to disease planets, ones they could cut ties with if they fail. Those still very early probes are: Zap-70 for autoimmune disease, cGAS/STING for autoimmune disease and cancer, SHP2 for cancer, PIM 3 for cancer and Factor XIIa for anticoagulation.
The process for development goes like this: Use AI to screen vast libraries of compounds for an effective molecule, preferably ones for diseases or assays that pharma companies have understood for some time but have had difficulty targeting. Test 100-300 of the computer’s results in a laboratory. Feed the lab results back into the computer. Test the computer’s new results. Repeat.
Collier said they’ve gone through one or two cycles for most of those probes. They’re aiming for a clinical trial by 2022, based on a timeline of a couple more rounds of screening and then a year of IND-enabling testing.
“That’s assuming success,” Collier said.
Like some other pharmaceutical artificial intelligence companies, Atomwise uses neural networks to identify compounds with the promise of speeding up preclinical work and cutting at the expanding costs, in dollars and time, of drug development.
It’s an idea that’s been around since at least the 1980s, but one that has attracted venture dollars as computing power has caught up to the vision. An incomplete list compiled from BenchSci, last updated this week, counts 167 AI drug discovery companies, including over 50 that, like Atomwise, try to generate novel drug candidates.
Still, these companies have yet to generate much clinical impact and one of the most high-profile companies, BenevolentAI, got taken down a peg and a billion dollars amid the Neil Woodford debacle. Writing up a recent high-profile study that claimed to use AI to find a promising new drug candidate in 21 days, Derek Lowe wrote the vaunted tech was coming – but be patient.
“The good news, though, is that there is no reason that virtual screening can’t do great things, eventually,” Lowe wrote. “We just have to get a lot better at it than we are now, and that’s as true as it was when I first heard about it in the mid-1980s.”