Three ways AI is changing drug discovery at AbbVie

What if the secret to finding cures faster lies not in a lab but in lines of code? This is the mission that fuels AI-driven drug discovery, a field where algorithms, data and analytics are being used to invent new and better treatments, and maybe even cures, faster. 

While AI tools like Chat GPT probably aren’t going to cure cancer, other types are already changing how new therapies are being discovered within pharma. And AbbVie is among those leading the way in creating and employing these new technologies at-scale. 

“The end game with AI is to bring new, more effective, treatments to patients in less time,” says Phil Hajduk, VP, IT information research, AbbVie.

Explore three ways AbbVie uses AI to broaden scientists’ understanding of disease and accelerate discovery of new medicines for patients. 

Today, inventing new therapies is both labor and time intensive, in part because drug discovery faces a major data problem. To find new drug candidates and anticipate how they will affect the human body, scientists have to aggregate and analyze vast and often disparate data, from genetic information, clinical trials, molecular interactions and multiple other sources. To do this alone would take scientists decades, or even a lifetime.

To solve this problem, and speed up the pace of discovery, researchers at AbbVie have built the AbbVie R&D Convergence Hub (ARCH), an industry-leading platform that centralizes and connects data from more than 200 internal and external sources. Coupled with machine learning (ML), ARCH can perform intelligent tasks, acquire knowledge and predict or classify data. 

One way ARCH accelerates drug discovery is by helping researchers identify new drug targets that play a causal role in disease.

Finding and validating targets early on in discovery matters because the quality of targets can impact clinical failure rates and drug approval rates down the line. However, this process is difficult because of how complex the biology of many diseases is, and how little is known about what drives these diseases.

Hard-to-treat diseases like cancer and autoimmune disease have many potential targets because they’re regulated by multiple molecules and intricate pathways. With AI and data integration tools like ARCH, scientists are now able to aggregate multiple pieces of knowledge about these diseases from many different sources, visualize patterns and connections among them, and conduct analyses on these data to figure out which genes or proteins they should pursue.

With a drug candidate chosen, scientists next focus on identifying and optimizing compounds that have the potential to be turned into future therapeutics. And in this step, AbbVie scientists actively utilize AI. 

When choosing the right drug candidate, researchers look for several attributes. Along with seeking compounds that are stable and effectively bind to their target, researchers also look for qualities such as good absorption, to ensure the drug can go to where it’s needed in the human body, or low viscosity, enabling the drug to be delivered via syringe.

To find a candidate that meets all these different requirements, researchers have traditionally relied on high-throughput screening of vast libraries of compounds – a process that requires considerable time and labor. But now AI methods are helping to accelerate this step.

With a form of AI known as large language models (LLM), researchers can engineer drugs in silico, meaning on a computer, rather than solely relying on high-throughput screenings. LLMs work by processing patterns from large datasets of known compounds, in the process ‘learning’ how differences in structure affect attributes like stability, viscosity, etc., and using those insights to predict new structures.

LLMs are more commonly associated with AI that looks at patterns in human language. However, these models also work for proteins like antibodies because, like English and French, proteins have a language of their own. 

In place of words or letters, proteins are represented by a string of 20 amino acids, and the order in which these amino acids are strung together define the attributes and properties of the protein. And just as generative AI uses patterns in the human language to predict the next most likely string of words within a sentence, protein language models can recognize patterns in the ‘language’ of proteins to predict new amino acid sequences. 

In drug discovery, LLMs are actively being used to predict new antibody structures that have desirable attributes, like low viscosity or high stability. To do this the traditional way, researchers would typically need to create different variations of the antibody, each with unique changes to the structure, and screen them all to find which versions were more stable or less viscous. But with LLMs, they can now find the best variations with less screening, making the process more efficient.

“What the AI model does, based on patterns within the data, is identify the ‘language’ or rules that control properties of proteins like antibodies. And with that you can go to it and say ‘I want an antibody that’s more stable or that binds well to this specific target’ and it’ll tell you the sequence that antibody needs to have that desired attribute,” says W. Blaine Stine, VP, discovery biotherapeutics, AbbVie. “That’s the power we’re starting to see emerge with these AI tools.” 

Yet another important aspect of drug discovery is ensuring therapies work for as many people as possible within the target population. Today, that science is imperfect: available therapies aren’t effective for every patient, and up until now, scientists have lacked the tools to understand why. But through the emerging field of precision medicine, which uses machine learning to develop therapies, researchers are getting closer.

AbbVie researchers use precision medicine to better understand the similarities and differences in genes, proteins and other potential biomarkers among patients affected by the same disease, and in turn use that knowledge to guide the development of therapies.

For instance, when analyzing clinical data on patients who responded to a given therapy compared to those that didn’t, precision medicine tools can reveal biomarkers that may be responsible for, or could have predicted, a patient’s response. Scientists can in turn take that knowledge back to discovery and use it to help them identify potential drug targets for that specific patient population. 

In oncology, it’s common to use biomarkers to understand patient response and stratify patients accordingly. But AbbVie is among those leading applications of precision medicine in immunology and other therapeutic areas.

“In the past two decades we have learned so much in oncology about the molecular biology of cancer and that has led to new treatments for patients,” says Heath Guay, Ph.D., executive director, precision medicine immunology, AbbVie. “We want to do the same for immunology because most of the therapies available treat disease broadly. We want to get to a point where we can tailor treatments to specific patient populations.”

Whether in precision medicine or drug targeting, all the ways in which AI is being used to help transform drug discovery signal that the future for patients will be nothing like the past.

“The discovery side has always been incredibly challenging, but with new capabilities like AI/ML, we now have the chance to be more ambitious in going after difficult-to-treat diseases, finding cures and delivering that patient impact we all hope for,” Stine concludes.