Roche and Genentech's latest AI collaborator raises $25M Series A to reinvent one of the oldest cancer targets
Jonah Kallenbach spent the summer before his senior year with Anton, the hulking fluorescent supercomputer housed in the midtown Manhattan offices of hedge fund D.E. Shaw. Armed with 512 processors running in parallel, it’s been used since 2008 to probe some of the biggest questions in protein folding and structure.
Kallenbach can’t talk much about what he worked on, but he was hooked, and after his last year at Harvard, he wanted to go into drug development. But when he went looking for pharma roles, he found few of the companies had space for computationals. Or they had space, but that space was small and cordoned off from the rest of a massive organization.
“No one was listening, no one was actually driving decisions,” Kallenbach told Endpoints News. That “was a lightbulb moment.”
It’s a familiar story by now for computational biologists, as was what Kallenbach did next: He founded his own company with a longtime college friend and collaborator, Ankit Gupta. Branded Reverie Labs, the Cambridge startup landed seed funding at Y Combinator and late last year a rare multi-target collaboration with Roche and Genentech. Now, they’ve raised their first significant capital, raising a $25 million Series A led by Ridgeback Capital to advance a pipeline of targeted cancer drugs.
The biotech will use machine learning and other computational methods to study the structure of kinases and ultimately design drugs that can knock them out. The proteins — among the first-ever hit with targeted cancer drugs — are an old focus for a young company, and Kallenbach says he initially faced pushback from investors.
“A lot of people were like, ‘Who cares about kinases?'” Kallenbach said. “Kinases are a solved problem, kinases are easy.”
Kallenbach argued, though, that kinases are an ideal focus for a computational startup. Because they’ve been studied voraciously over the past few decades, researchers have developed huge data sets on their structure and function — data sets that machine learning can turn into druggable insights.
Kinases are also an area where specificity becomes crucial; a “healthy” kinase and an oncogenic one may only differ by an amino acid or two, and a drug that hits both may run into the kind of safety issues that plagued the Pi3K space for a decade. New tools can help discern the difference.
Kallenbach and Gupta hope that for their targets, they can truncate the years of work and setbacks that went into the first Pi3K approval. Already working with Roche, they want to eventually look something like Nimbus: A sought-after discovery engine that can develop molecules partially in-house, before large companies bring them through the home stretch.
The engine will be particularly important, Kallenbach said, as developers push deeper into combination approaches that overcome resistance to individual drugs. Every new drug in a combo brings the potential of new toxicities, raising the bar for how safe each individual molecule is.
“The future of cancer therapy is to actually be able to cure people of cancer, and not just give them a 3 or 6-month boost in PFS,” he said. “And the only way we can actually do that and have the patient tolerate that therapy, is if those therapies are pretty specific.”