What do we really mean by “undruggable” diseases?
And how AI drug creation can take us from “undruggable” to “not-drugged-yet”
By Andreas Busch, Absci Chief Innovation Officer
The way we bring drugs to market today is slow, expensive, and inefficient. On average, it takes more than ten years and costs more than a billion dollars, and about 95% of drug candidates fail along the way. The costs and risks are so high that drug companies tend to focus on big, easy-to-treat diseases.
That’s bad news for patients suffering from one of the world’s many “undruggable” diseases — conditions for which there are currently no effective drugs or therapies. These diseases involve challenging targets with complex biological mechanisms that are difficult to solve with existing tools. Examples of well-known diseases traditionally considered undruggable include Alzheimer’s disease, Parkinson’s disease, and certain types of cancer.
But diseases can’t simply be categorized as druggable or undruggable. Some disease targets may face one or two key challenges, while others may face significant challenges in several aspects, making them less druggable. Druggability exists on a continuum from completely cured diseases all the way to the most challenging, yet-to-be-drugged targets. And when you add in additional factors that have nothing to do with being able to target a specific disease — manufacturing feasibility, patient market size, regulations, and intellectual property, to name a few — the concept of druggability can take on a broad, hard-to-define social meaning.
So what makes a disease undruggable? There can be many different reasons:
- Researchers haven’t found a good target for it — a protein involved with modifying the disease that we can bind a drug to in order to treat the disease.
- Many diseases are of multi-factorial origin and addressing a single target may not be sufficient to have the needed impact.
- The disease target is located in a part of the body that is difficult to reach, like the brain.
- The drug treatment itself causes toxicity or side effects, either by creating an unwanted metabolic process or by the compound’s direct effects.
A needle in a molecular haystack
To overcome some of these challenges, drug discovery has shifted some of its focus on small molecule drugs to biologic drugs, including antibodies. Whereas small molecule drugs are chemically synthesized compounds, biologics like antibody therapies are large, complex molecules produced using living cells and are designed to target specific proteins or cells in the body. Antibody therapies offer several advantages over small molecule drugs, including greater specificity and fewer off-target effects, opening up new and improved therapeutic possibilities for patients.
But finding the right antibody therapy for a disease can be a tall order. Some estimate that only 10-15% of human proteins are disease-modifying, and only 10-15% of human proteins are druggable. If correct, that means that only about 1%-2.25% of disease-modifying proteins are likely to be druggable. Discovering new drug candidates with traditional approaches is a lot like finding a needle in a vast molecular haystack.
Enter AI drug creation
Artificial intelligence (AI) promises to impact medicine in many ways, from helping us to understand the underlying biology of diseases to even tailoring treatments to a patient’s specific genome. When it comes to finding new antibodies to bind to disease targets, AI is transforming the pharma paradigm from drug discovery to drug creation. Instead of looking for a needle in a haystack, AI drug creation makes the needle.
How? With generative AI — the same kind of technology that underlies AI chatbots and image creators. Using vast amounts of biological data created in our wet labs, machine learning algorithms can now begin to understand the fundamental rules determining how drugs effectively bind to targets. This enables the algorithms to design new antibody therapies that may never be found in nature.
In short, AI goes beyond traditional trial-and-error approaches to significantly reduce the time it takes to identify a clinical drug candidate with a higher probability of success.
Pushing targets to the druggable end of the spectrum
AI drug creation has the potential to increase the “druggability” of various diseases, pushing them from the “undruggable” end of the spectrum to the more “druggable” extreme. One example of this has been demonstrated in HER2, or human epidermal growth factor receptor 2. This protein is overexpressed in certain types of cancer, such as breast and gastric cancer. The overexpression of HER2 can lead to uncontrolled cell growth and division, making it an attractive target for cancer therapy.
In the past, HER2 was considered an “undruggable” target because of its complex structure and the lack of suitable drug-binding sites. In the 1990s, researchers developed an antibody called trastuzumab (Herceptin) that could bind to HER2 and prevent its activation. As the first FDA-approved therapy targeting HER2 for breast cancer, trastuzumab has improved the prognosis for thousands of people with HER2-positive cancer. And as subsequent HER2-targeting drugs have shown, there’s still room for improving this breakthrough therapy.
In August 2022, Absci published details showing how its AI drug discovery platform rapidly optimized multiple parameters important to drug development for the trastuzumab antibody. Absci was not only able to accurately design and predict an antibody’s binding affinity given its amino acid sequence, but it was also able to optimize antibody designs for a high degree of “Naturalness” — meaning they may have a lower likelihood of triggering unwanted immune responses.
In January 2023, Absci went a step further. Its AI platform designed and validated new antibodies for HER2 and other targets using zero-shot generative AI, a method that purposely excludes data about existing antibody drugs like those above when training the AI algorithms. Specifically excluding such data generated antibody designs that were unlike those found in existing antibody databases. The entire zero-shot process involved millions of AI designs and billions of measurements, and the entire process — including experimental validation of the AI-designed antibodies in the laboratory — took only about six weeks.
The takeaway: AI drug creation is poised to speed better antibody treatments into the clinic and ultimately to the patients who need them, pushing targets to the druggable end of the spectrum.
The tipping point in druggability
By speeding preclinical development times and increasing clinical rates of success, AI drug creation, as envisioned by Absci, will decrease the risk and cost of pursuing not-drugged-yet diseases. As AI drug creation matures, it could help the pharma industry reach a tipping point where it becomes increasingly feasible to reach beyond the low-hanging fruit of the biggest drug markets to a wide range of less prevalent diseases. When that happens, it could give new hope to rare disease communities and even signal the beginning of more personalized medicine.
AI drug creation has more work to do before bringing its first therapy to the clinic. But researchers at Absci are developing new approaches to help overcome these challenges every day. We believe AI drug creation will enable us to uncover potential therapies for more and more diseases with the click of a button. When that happens, it will represent a new era for not-drugged-yet diseases.