Artificial intelligence (AI) has yet to transform the drug development process, but some efforts seem more promising than others.
On Monday, cancer research giant Memorial Sloan Kettering and life sciences AI pioneer Absci announced a first-of-its kind partnership to discover six novel therapies for cancer using generative AI, promising to bring new drugs to clinical trials next year.
Also: How AI hallucinations could help create life-saving antibiotics
MSK will identify the target on cells to go after, and Absci will use its gen AI to create a de novo antibody that will bind to that target.
“This is the first collaboration of this kind that Absci has done, especially with an institute like MSK,” said Sean McClain, founder and CEO of Absci, in an interview with ZDNET. “It provides a really great synergy: the knowledge and expertise that MSK has in oncology, and these novel targets that they’re going to be bringing forward, with Absci’s ability to design exciting drug candidates with our AI platform.”
<!–>
–>
“If MSK does bring a novel GPCR, our platform is really well suited for that, to be able to find an antibody that combines with that target,” what McClain calls “helping create the biology.”
Absci
The imprimatur of MSK is a major vote of confidence for the very young world of life sciences AI.
“We are always looking for new ways to push things forward for patients all around the world, and AI is clearly an area where we need to be involved,” said Dr. Gregory Raskin, MD, senior vice president of Technology Development at MSK.
Also: Hooking up generative AI to medical data improved usefulness for doctors
“We have never before partnered to make new antibody drugs with a company that focuses on AI,” he said. “We hope to be not just a leader in cancer care, but a leader in cancer care AI at MSK.”
The collaboration is described as a “co-development” agreement and a “50/50 partnership,” with both parties funding the initiative, although funding amounts were not disclosed.
Talks between MSK and Absci began at the J.P. Morgan Healthcare conference in San Francisco in January, said Raskin, and have evolved over the ensuing seven months.
The division of labor involves MSK coming up with a target, which the two parties will then discuss and agree upon, and Absci devising a design for an antibody, or a series of antibodies, against the target.
In addition to the computer simulations Absci can run, and its own wet lab facilities, MSK will help with “core facilities and scientists in our institution that are world experts in determining whether a drug is going to be able to defeat a tumor and be safe,” said Raskin.
“Once we have the target identified, we will then be using the generative AI models to design the antibodies towards these targets to achieve the biology,” said McClain.
Also: How GenAI got much better at medical questions – thanks to RAG
“MSK boasts over a hundred research labs with people working on oncology, looking for new cancer targets,” said Raskin. He noted that 13 drugs that have been approved by the U.S. Food and Drug Administration (FDA) were invented at MSK, including Danyelza, a treatment for pediatric neuroblastoma, in 2020, and Erleada, a treatment for non-metastatic castration-resistance prostate cancer, in 2018. The FDA is responsible for approving clinical trials and either accepting or rejecting drugs.
<!–>
–>
“In the case of mouse antibodies, it’s a time-honored approach, and it’s time-consuming, and labor intensive,” observed Raskin. “You might get a bad run of antibodies that don’t bind well to your target.”
“We hope that this method is going to be faster getting into our patients — that’s just key.”
Memorial Sloan Kettering
MSK has its own patient population that it can use to test drugs that may come out of the partnership, Raskin told ZDNET. The hospital runs around 1,800 clinical trials, some for outside parties, some for internally-developed drugs.
“We have the ability to write our own INDs, and we can start trials in our own patients with these technologies,” he said, referring to the “Investigational New Drug” filings necessary to submit to the FDA.
The appeal of AI, said Raskin, is the potential for the technology to speed drug development that takes, on average, a decade. By using gen AI, new drugs can be rapidly conceived and simulated on the computer — in some cases shaving years off of the typical process of in vitro chemistry and in vivo animal tests.
“In the case of mouse antibodies, it’s a time-honored approach, and it’s time-consuming, and labor intensive,” observed Raskin. “You might get a bad run of antibodies that don’t bind well to your target.”
“We hope that this method is going to be faster getting into our patients — that’s just key.”
Also: Generative AI could lower drug prices. Here’s how
For a first-ever partnership by Absci with a hospital on novel drug targets, said McClain, the important thing is “being able to take the drugs that we co-develop together into patients at MSK, and being able to have MSK really help on the translational side, making sure we got the right clinical strategy.”
Absci already has partnerships with multiple pharmaceutical giants, including AstraZeneca, Almirall, Merck, and AI chip titan Nvidia.
McClain said the MSK collaboration is different in that the institute is a nonprofit, as opposed to a Big Pharma for-profit operation. Because the commercialization of therapies can be enormous, Absci and Memorial Sloan Kettering plan to bring in a pharmaceutical partner to ultimately commercialize any drugs, ideally after demonstrating a “proof of concept” on their own.
As for what targets the six drugs will go after, “I don’t think we know yet,” said Raskin. “We have to talk with our scientists and see what matches with what Absci thinks.”
Finding a cancer target is itself an intense task, noted McClain. Absci has resources to formulate novel antibodies from gen AI, but the company needs the expertise of scientists trained in hunting through the body’s drug receptors for worthwhile places to strike.
“A lot of GPCRs are emerging as novel targets,” he said, referring to “G protein-coupled receptors,” which are the largest family of receptors that are targeted by approved drugs.
“If MSK does bring a novel GPCR, our platform is really well suited for that to be able to find an antibody that combines with that target,” what McClain calls “helping create the biology.”
Also: FDA approves first prescription-only app for depression
Given there has not been substantial clinical evidence to date of AI’s usefulness, the collaboration, should it produce definitive clinical data, could be an important proof. “There are some small-molecule companies out there that will have Phase II [clinical trial] readouts,” said McClain. “But in terms of antibodies, these will be some of the first that hit the clinic.”
Absci has shown early proof that gen AI can design novel antibodies that bind to cancer targets. In March, Absci reported development of novel antibodies against what’s called “human epidermal growth factor receptor 2,” or HER2, a human oncogene that has been linked to some forms of breast cancer.
The AI model had not been fed data on existing, successful antibodies against HER2, nor had it been given explicit information about how to successfully attach — or “bind” — to HER2.
Absci’s lead drug candidate in its pipeline, ABS-101, is a treatment for irritable bowel disease (IBS). The novel antibody, developed using gen AI, binds to the TL1A protein in immune cells whose over-expression has been linked to a variety of inflammatory autoimmune diseases. The antibody was developed from scratch in just 14 months and at a cost of less than $5 million, McClain emphasized.
<!–>
ABS-101 is expected to begin Phase I clinical trials next year. Another project, ABS-301, is an undisclosed “immune-oncology” target that has been validated by Absci internally.
“You are starting to see these AI-generated antibodies and small molecules make it into the clinic,” said McClain.
Given concerns over the privacy of patient data, it’s important to maintain separation between MSK’s patient data and Absci’s AI model training.
“We’ll use their data and expertise to select the cancer target to go after, and then we’ll use our model to genre the antibody,” said McClain. “We’re not planning on using MSk data to train our models, we’re going to be generating that data in-house, using that for training – so, it’ll be completely separate and firewalled from that perspective.”
