AI is being used everywhere, including in drug discovery. There are many things AI drug discovery technology can do to further advance treatments for various diseases, but there are some things that AI cannot do. Understanding the applicability of AI predictions is essential. Without that understanding, it is easy to believe that AI alone can solve any problem autonomously when, in fact, researchers are also essential to the process.
AI drug discovery is vital to the future of the pharmaceutical and healthcare industries. Machine learning, an AI technique, is advancing the discovery of new pharmaceuticals quicker, cheaper, and more effective. Importantly, AI can pinpoint previously unknown causes of diseases while accelerating treatments for patients with specific biological profiles.
AI’s successful use depends on the quality of decisions, such as which target proteins to tackle, which compounds to advance, and how to conduct clinical trials using clinical efficacy and safety-related endpoints. One formidable challenge is that the data available may be inadequate to ensure these decisions, while experts’ decisions might contain biases.
This is one reason why pharmaceutical companies partner with AI drug discovery companies.
Drug discovery challenges and how to solve them
The biggest challenge for pharma companies is that clinical studies are capital- and time-intensive and are riskier than ever. Fewer “low-hanging fruits” remain, so a higher science level is necessary, even in the early drug discovery stages.
AI is ideal for unbiased drug discovery because, given a target, it can sift through a staggering amount of molecules and compounds, weeding out those that do not qualify without any human bias. It can also represent general and specific chemistry principles from data-centric learning models. Specifically, it finds complex relationships from an unorganized data collection and discovers unknown areas in the biological and chemical space, generating plausible and validatable hypotheses for early drug discovery projects.
It is important to think critically about AI
There has been considerable noise about ChatGPT recently, both positive and negative. OpenAI, the creator of ChatGPT, is working on artificial general intelligence (AGI), though today’s state of the art is artificial narrow intelligence (ANI). ANI can effectively solve a specific problem like drug discovery or contract review, but it can’t be used to solve a different problem, such as recommending product upsells and cross-sells.
Those unfamiliar with how AI works may believe it can solve any problem; even scientists may have this view.
Generative machine learning models for novel molecule design require the context of research because each case will have distinct requirements, which means the current generative AI in chemistry is still an ANI. Without a proper understanding of this limitation, one might think AI will elect the molecular candidates with the highest probability of success when AI’s choice differs from human experts, which is not always true.
Where AI fits in the drug development cycle
The success of any drug discovery project relies on selecting the best target protein for a given disease, while the best can be defined in various ways. AI can assist researchers in this regard. For example, AI can provide a list of novel and promising target proteins with some clues about the biological mode of action and hypotheses that experiments can validate. Since AI is scalable – meaning it can suggest as many scenarios as the researcher wants, in most cases – it can provide alternative options, translating to a substantial strategic advantage.
The most foundational limitation of AI is the lack of high-quality data for predictive models. For example, the number of molecules experimentally tested on a novel target protein is usually less than one hundred, while the number of all known molecules is bigger than one hundred million, and billions of compounds can be purchased. This ratio means the validation of AI prediction in an unexplored space is challenging, which limits AI’s extrapolation in molecular design. In fact, early drug discovery is a series of many distinct tasks with similar limitations, making the process far more complex. So, most AI tasks and workflows (different tasks in different orders) in the drug discovery process must be validated by well-designed experiments.
Another challenge is teaching AI the basic concept of science. While AI can understand the data represented, all the context or implicated principles should also be appropriately represented, which is fundamental scientific work. Without a sophisticated learning process, AI models can make many mistakes that seem silly to people. This is why AI drug discovery needs a balance between AI and drug discovery experts to maintain its applicability to real-world drug discovery projects.
How pharmaceutical companies are approaching AI drug discovery
Considering the strict data policies, pharmaceutical consulting companies cannot simply outsource drug discovery to a third party. What they can do is collaborate with partners that have AI-based drug discovery capabilities. One benefit to this approach is that those partners understand what it takes to implement the technology successfully, which involves people, processes, and technology. One easy way for pharmaceutical companies is to replace a single task with an AI-assisted one as it is more straightforward to compare two tasks, though it is more conservative with less impact. Adopting AI for drug discovery for a profound productivity gain requires organizational changes and a long-term plan for human-machine partnerships.
While AI can accelerate the time required for drug discovery, it can only be effective with human oversight and judgment. Researchers at pharmaceutical companies and AI drug discovery solution providers must work together to ensure the best possible outcomes.
