Empress library illustration
Empress Therapeutics, which exited stealth mode a year ago, has already discovered 15 drug leads across multiple diseases and target classes. Launched in 2020, the Flagship Pioneering startup is now driving towards filing multiple IND applications within the next 24 months.
While AI has played a role in compressing those timelines, the company is focusing on more than efficiency gains. “There are two aspects to how you make drug discovery better, and AI does have a role to play. Speed and efficiency is one part, but predictability and certainty is probably the better area to focus on,” said Jason Park, CEO and co-founder of Empress Therapeutics.
Chemilogics: Combining patient data, DNA and AI
Empress Therapeutics employees in the lab
Empress shares Flagship Pioneering’s core philosophy of looking to evolution and genetics to drive therapeutic breakthroughs. Similar to how Moderna tapped the information molecule, mRNA, to program cells for protein production, Empress has developed a platform known as Chemilogics that combines patient data, DNA, AI, and synthetic biology to help pinpoint small molecule, or chemistry-based, drug candidates.
While interest in large molecule drugs has increased in recent years, small molecules have a number of advantages. Their inherent properties make them uniquely suited to cross cell membranes, reaching intracellular targets that are often inaccessible to larger molecules. In addition, their potential for oral bioavailability, makes them a cornerstone of modern medicine.
Tackling the challenge of small molecule discovery
Jason Park, Ph.D. [Empress Therapeutics]
Instead of grappling with the astronomical number of potential drug-like compounds, Empress focuses on the rich chemical diversity already encoded within the human metagenome, which spans the individual’s genome as well as the collective genomes of the microorganisms residing in and on the human body. The human genome contains about 20,000 protein coding genes. Conversely, recent estimates suggest that there are more than 170 million unique protein sequences in the gut microbiome alone.
Mining the human metagenome for drug candidates
The rich diversity of protein sequences within the metagenome means a wide variety of enzymes with potentially novel functions. “Another way of thinking about this — basically every chemical compound in your body, except for the ones you’re eating and taking in through that route, is touched by enzymes when they’re made or modified,” Park said.
Empress thus exploits genetic data to explore how DNA programs cells to produce chemical compounds, via enzymes, allowing it to identify small molecules that have evolved within the human body to interact with specific disease targets. “If you picture the central dogma — DNA encodes RNA encodes proteins.” Empress set out to shed light on the role of enzymes in catalyzing chemical reactions. “”The big question was, what if you could start to predict how compounds in your body are ultimately tied back to the genetic sequence? Can you go from DNA to chemistry?”
From DNA to chemistry
To that end, the company uses techniques such as natural language processing to read DNA and investigate the role particular sequences play in encoding multiple enzymes that play a role in producing molecules of interest. Nature has already vetted such molecules over millennia.
“We know the parent compound is already inside the human body, so there’s reason to believe it’s safe,” says Park. “And we know if it’s conserved, it’s dysregulated in disease, it’s encoded in genetics — that gives us a lot of confidence and certainty that this molecule is important.”
A convergence of technologies and talent
The company’s strategy to mine the metagenome to explore the “most privileged set of chemistries” wouldn’t have been possible a decade ago, Park said. That’s because the company relies on the convergence of several elements. “You need the data,” Park emphasizes. The recent explosion of metagenomic data provides powerful fodder for AI algorithms. “But you also need things like synthetic biology, advances in NLP, genome mining, causal discovery, causal inference tools,” Park said.
“It’s similar to what we saw in the tech industry,” Park said, referring to the convergence of technological streams — breakthroughs in algorithms including the emergence of transformer models, GPU advances, the availability of elastic cloud storage and processing as well as the collaboration of multi-disciplinary experts driven by an entrepreneurial mindset.
This convergence is embodied in Empress’s own leadership team, which includes industry veterans like Chief Scientific Officer, Murray McKinnon, who brings more than 35 years of experience in drug discovery and early development. McKinnon has contributed to the development of numerous blockbuster drugs, including Orencia, Nulojix, Remicade, Simponi, Stelara, and Tremfya.
Questioning limits with experiments
This blend of experience and a willingness to challenge conventional wisdom is at the heart of Empress’s approach. “We’ve got entrepreneurs, a number of scientists who know that what we call scientific dogma is really just a veneer for a set of experiments,” Park said. “We’ve got entrepreneurial scientists who maybe have some startup experience or came out of academia. They don’t really know what the limits are.”
“We have that entrepreneurial spirit of ‘we can do anything,’ the platform capabilities to make drug discovery repeatable and an engineerable problem rather than just a scientific endeavor, and then people who know what a good drug should look like,” Park said. “That’s part of what’s fun about working in a small biotech. You bring all those people together and focus on the mission of making great medicines, putting everything in place to increase the certainty and speed of doing that. Because if we’re right about this, and the data look good so far, we’re not going to make just one medicine — we’re going to make drug after drug after drug.”
Filed Under: clinical trials, machine learning and AI
