Unlocking the potential of AI in drug discovery: Guest Commentary

The combined power of data and people are needed to realize the benefits of AI in drug and target discovery

2023 is likely to be remembered as the year artificial intelligence went mainstream, inspiring excitement about its potential across virtually all sectors of society. And while the recent surge in consumer adoption of generative AI tools has been a catalyst for their acceptance, the biotech industry had already been exploring their utility for years. What’s become clear is that AI is positioned to make major contributions to drug design and target identification, if we can leverage it correctly — which will take the combined power of digitally-enabled people and the technology itself.

Among the potentially far-reaching applications of AI in healthcare, drug discovery is one of the most critical. Not only is drug discovery the powerhouse that fuels creation of innovative medicines for patients, but the conventional process of drug discovery is long and notorious for its high attrition rate. For instance, across all therapeutic areas in the industry, 90% of drugs fail clinical trials — a number that rises to 97% when looking at cancer drugs.

Researchers have always sought ways to improve the drug discovery process, of course in terms of speed, but certainly also in terms of quality, probability of success or even unlocking whole new approaches. This includes using increased adoption of automation techniques to test large libraries of new molecular entities, synergy between hypothesis-driven research and large-scale genomics, genetics and molecular design in the very early stages to identify potential drug candidates. Although the industry has made progress, the high failure rate remains a challenge.

AI promises to make a material dent in that failure rate. 

I started my nearly 20-year healthcare journey as a computational oncology biologist. I joined Genmab, in part, because of its strong scientific fundamentals in antibody technology, and was presented with an opportunity to establish an embedded discovery data science team to integrate data and AI/machine learning close to the company’s core expertise in antibody biology.  

This experience has made me a firm believer that silos must be broken if companies want to reap the benefits of AI. Integrated multidisciplinary teams with data scientists in close partnerships with other functions are highly effective. We’ve found that by ensuring our data scientists are not only fully embedded across discovery, but also clinical development and commercialization, we can be better equipped with insights to make decisions that can ultimately speed our progress. At the same time, organizations can’t only rely on data scientists alone — it is essential to cultivate a data savvy culture, a high-throughput mindset, and an affinity with AI across the organization.

With the right team and the right technology, organizations can position themselves to deliver innovations to patients faster and smarter — starting with drug discovery. I’m also more convinced than ever that two applications within drug discovery have transformative potential for the pharmaceutical industry and, ultimately, patients.

1)  Learning the language of proteins:

Proteins play a key role in drug discovery as they are often the targets upon which drug molecules exert their therapeutic effects. Understanding proteins — their functions, structures and interactions with other molecules — is crucial for identifying effective drug targets. Likewise, learning the interplay between proteins and drug molecules is essential for developing new therapeutic agents and getting insight into the disease mechanism at a molecular level.

Just as chatbots learn the rules and patterns of human language, AI can be used to learn the language of proteins. Like our alphabet, proteins can be described as a language with the 20 natural amino acids spelling the words, or ‘tokens’ in the AI vernacular. As antibody molecules and their related cousins are also proteins, using this language, we can also gain a deeper understanding of the function and structure of potential drug molecules.

This journey is only just starting. Currently, Genmab and other leading companies in the industry are using AI and language models to characterize and predict features of drug candidates very early in the process. This enables us to make better choices early on, as well as more informed choices about molecules we want to advance as clinical candidates. However, to fully enable the power of these AI approaches, they need to be fed with data. Across the industry, there is a clear understanding that we need to generate more unbiased data that will help us better train and develop the right AI models that will unlock the future potential of AI. Genmab is committed to this challenge.

2)  Unlocking drug targets:

Another field presenting great opportunities for AI and digital technologies is the target discovery area.

The typical process of identifying biological molecules and pathways with therapeutic potential can be artisanal and time-consuming. However, with AI and data science tools that can analyze large datasets, researchers can gain a better understanding of diseases and identify promising targets earlier, saving time and resources.

At Genmab, we’re setting up new ways to identify targets, moving away from the more traditional “trial-and-error” approach. We integrate large-scale data analysis of genomics data, high-throughput screening approaches and innovative new in vitro systems to get in-depth biological knowledge. This allows us to pull AI/digital approaches, next-generation lab-based systems and high-throughput screening approaches into an integrated closed-loop process that scales up both the identification and validation of targets. AI plays a huge role in the processing, integration and understanding of the vast amounts of data that result from this way of working.

Charting the path forward

AI applications are already revolutionizing the way we discover and develop novel therapies, and we’ve only just begun to tap into their potential. As an industry, we find ourselves at a crucial juncture, confronted with both challenges and opportunities brought about by this powerful technology. One key point to reiterate is that data fuels all of this, and how we deal with data and set up ways to generate data at scale is key to being successful. One important note is that we cannot get there by only leveraging the legacy data of an organization for AI, it is imperative to generate data specifically for AI use cases.

Most importantly, success in fully unlocking the potential of AI will rely on people and how we organize. Traditional siloed organizational structures will not effectively bring the right components together — integrated multidisciplinary teams operating from a strong core of computational, technological and biological expertise will be crucial to success. The individual quality of scientists and data scientists and the way organizations set themselves up to collaborate are what will empower them on their journey of innovation, drug discovery and beyond.

Chris Klijn is VP, head of discovery data science & target discovery at  Genmab A/S.

Signed commentaries do not necessarily reflect the views of BioCentury.