AI-based technologies are increasingly being used for things like virtual screening, physics-based biological activity assessment, and crystal structure prediction of drugs.
Despite the buzz around artificial intelligence (AI), most industry insiders know that the use of machine learning (ML) in drug discovery is nothing new. For more than a decade, researchers have used computational techniques for many purposes, such as finding results, modeling drug-protein interactions, and predicting reaction rates.
What’s new is the hype. As AI has taken off in other industries, countless start-ups have emerged promising to transform drug discovery and design with AI-powered technologies for things like virtual screening, physics-based biological activity assessment and crystal structure prediction of drugs.
Investors have made huge bets on the success of these start-ups. Investments reached $13.8 billion in 2020 and more than a third of big pharma executives say they use AI technologies.
Although a few “native AI” candidates are in clinical trials, around 90% remain in discovery or preclinical development, so it will take years to see if the bets pay off.
With big investments comes high expectations – medicate the must-haves, dramatically shorten lead times, virtually eliminate wet lab work. Insider information projects that discovery costs could be reduced by up to 70% using AI.
Unfortunately, it’s just not that easy. The complexity of human biology prevents AI from becoming a silver bullet. Furthermore, the data must be abundant and clean enough to be used.
Models must be reliable, potential compounds must be able to be synthesized, and drugs must pass real-world safety and efficacy tests. While this harsh reality has not slowed investment, it has led to fewer companies receiving funding, devaluations, and the halting of some nobler programs, such as IBM’s Watson AI for the discovery of medications.
This begs the question: is AI for drug discovery more hype than hope? Absolutely not.
Do we need to adjust our expectations and position to be successful? Absolutely yes. But how?
Three keys to implementing AI in drug discovery
Implementing AI in drug discovery requires reasonable expectations, clean data, and collaboration. Let’s take a closer look.
1. Reasonable Expectations
AI can be a valuable part of a company’s broader drug discovery program. But, for now, it’s best seen as an option in a toolbox. Clarifying when, why and how AI is used is crucial, albeit difficult.
Interestingly, investments have largely fallen to companies developing small molecules, which lend themselves to AI because they are relatively simple compared to biologics, and also because there are decades of data to build on. models. There is also wide variation in the ease of applying AI across discovery, with models for early detection and prediction of physical properties appearing to be easier to implement than those for predicting targets and assessing performance. toxicity.
While the potential impact of AI is incredible, we must remember that good things take time. Pharmaceutical technology recently asked its readers to predict how long it would take for AI to reach its peak in drug discovery, and by far the most common answer was “more than 9 years”.
2. Clean data
“The main challenge in creating accurate and applicable AI models is that the available experimental data is heterogeneous, noisy and sparse, so proper data curation and collection is of utmost importance.”
This quote from a 2021 Expert opinion on drug discovery this article speaks beautifully about the importance of collecting clean data. Although it refers to ADEMT and activity prediction models, the statement is also generally true. AI requires good data, and lots of data.
But good data is hard to come by. Publicly available data may be inadequate, forcing companies to rely on their own experimental data and domain knowledge.
Unfortunately, many companies struggle to capture, federate, leverage, and prepare their data, perhaps due to skyrocketing data volumes, outdated software, incompatible lab systems, or disconnected research teams. Success with AI will likely elude these companies until they implement technology and workflow processes that allow them to:
- Facilitate error-free data capture without relying on manual processing.
- Manage the volume and variety of data produced by the different teams and partners.
- Ensure data integrity and normalize data for model preparation.
Businesses hoping to leverage AI need a complete view of all their data, not just fragments. This requires a research infrastructure that enables IT and experimental teams to collaborate, unite workflows, and share data across domains and locations. Careful standardization of processes and methodology is also necessary to ensure that the results obtained using AI are reproducible.
Beyond collaboration within organizations, major industry players are also collaborating to help AI reach its full potential, making security and privacy key concerns. For example, many big pharma companies have partnered with startups to help boost their AI efforts.
Collaborative initiatives, such as the MELLODDY project, have formed to help companies leverage pooled data to improve AI models, and vendors such as Dotmatics are building AI models using the collective experimental data of customers. .
About the Author
Haydn Boehm is Director of Product Marketing at Dotmatics, a leader in scientific R&D software that connects science, data, and decision making. Its enterprise R&D platform and science-preferred applications drive efficiency and accelerate innovation.