Putting the AI train on the clinical track to success
Amer-Denis Akkad connects the dots between AI drug creation, translational research, and clinical developability.
Despite decades of biotech breakthroughs and the Warp Speed innovations that came with COVID, the cost of bringing new drugs to market is going up, not down. And most of the eye-popping $6 billion average cost is due to failures in late-stage development, suggesting the drug development industry isn’t great at picking winners from the very start. How can AI drug creation increase our clinical probability of success — and do it faster than current routes?
That’s where people like Amer-Denis Akkad come in.
“We have powerful new AI capabilities to design antibodies, but we need to think ahead about how we use it to lay the tracks for the train to run to the desired destination without delays, with each stop along the way representing a milestone in the drug creation value chain,” says Denis. “That’s where I see my role: putting the AI train on the clinical track to success.”
As Absci’s Head of Translational Research, Denis brings his experience at Bayer to help harness AI drug creation in efforts to design safe, effective, and developable drugs that meet the real needs of patients. It’s a complex problem that takes a seasoned pharma team, a scalable and integrated wet lab, and dynamite AI all working together to solve the problem.
Drug failure and the translational challenge
To succeed, drugs must meet a long list of requirements. For starters, they have to be safe and effective, they must have convenient and effective dosing, they can’t cause bad side effects, and they must be affordable to make and distribute. Translation lies at the center of meeting these requirements by bridging innovation in the lab with results in the clinic.
“Translation has many different aspects,” says Denis. “One is, how does your disease hypothesis translate from your models to the patient? Another aspect is, how does your project move across different stages of drug development?” These two different aspects come together when also considering the regulatory and commercial landscape. By accounting for these broad aspects starting in the early preclinical stage, Denis says you’re already improving your odds of success.
Why do drugs fail? The $6 billion question.
|
Drugs can fail for many reasons on their way to the clinic, including lack of clinical efficacy, unmanageable toxicity, poor technical properties, and/or an overall lack of commercial need. Often, the root cause of the lack of clinical efficacy is that we get the starting point wrong, namely the incomplete understanding of the underlying disease rationale for the particular indication or even individual patient.
For example, a drug that is designed to inhibit a particular enzyme may be effective in animals but is ineffective in humans because the enzyme is expressed differently or encounters compensatory or escape mechanisms. Or, the drug may interact with other molecules or pathways in humans in unexpected ways.
“Your disease rationale may be flawed from the beginning so that yes, it might work in animals, but in actuality, you’re only focused on a small aspect of the disease,” says Denis. When that happens, “your drug doesn’t have an impact in the grand scheme of the disease biology.”
Increasing clinical probability of success
AI drug creation may help avoid these kinds of costly failures by generating better drug candidates from the outset.
“But that’s just the beginning of the journey,” says Denis. Bringing drugs to market requires understanding patient needs and the pharmacology of their disease, then applying that knowledge to all different stages of experimentation. “With a well-integrated wet lab and AI,” he says, “we can generate a wealth of understanding to tackle hurdles across the value chain.”
Increasing clinical probability of success (PoS): Key opportunities in AI drug creation
|
Among the potential ways AI drug creation can improve the clinical probability of success (PoS), it can help make smarter trade-offs in early drug discovery when optimizing among many parameters of interest.
“With multiparametric lead optimization, we have the ability to potentially make much better drugs with more favorable risk-benefit assessments,” says Denis. On one end of the spectrum, he points to chronic, non-life-threatening conditions like psoriasis or arthritis. Here, drug makers aim to be as safe and effective as possible, while optimizing for dosing that is better than what’s already on the market – say, one pill a day instead of two pills twice a day. On the other end of the spectrum, Denis points to life-threatening diseases like cancer that demand acute treatment. There, for example, patients might tolerate adverse effects to some extent in exchange for a drug that helps to overcome an immediate life-threatening disease. “AI drug creation helps drug makers consider these factors much more rationally,” he says.
In AI drug creation, you can also apply the same multiparameter optimization approach to properties that are important to CMC, manufacturability, and developability. Self-interaction, polyspecificity, hydrophobicity, thermal stability, compatibility with high-concentration formulations, and resistance to stress challenges are some of the parameters that drug manufacturers can consider early in drug design to increase the chances of progressing all the way through the drug development process.
“As the wet lab data grows and grows, AI drug creation streamlines how we architect the development path from in vitro studies through to animal models and all the way to clinical trials,” Denis says. Using knowledge graphs and other forms of multidimensional data integration, Denis says we are able to work in an infinite design space over very short periods of time.
“For us,” he says, “the only real limit is antibody data, and Absci’s roots in cell line development give us a good head start with an integrative approach of wet lab to AI activities.”
Translating innovation into clinical success
AI drug creation is poised to improve the clinical probability of success (PoS) of new drugs. By generating designs that better address the disease rationale, identifying better targets, and optimizing drug candidates for multiple parameters, Denis thinks it can help drug developers avoid costly failures and bring more effective and affordable therapies to patients faster.
Denis says drug discovery is a tough business, and that there are challenges ahead. But the field is moving incredibly fast, and he’s optimistic that de novo generative AI will change the industry and make a huge impact on patients.
“Absci is on another level of disruptiveness,” he says, “combining technologies that can identify new targets, new leads, and which patients will benefit the most. Putting that all together, with the experts and talent that Sean and the team have brought together, is a really important part of the journey for me.”