There are many applications of AI in drug discovery. In this article we have chosen to focus on four to illustrate how the framework can be used to assess AI maturity.

1. Repurposing existing drug candidates:

One of the most effective applications of AI in drug discovery is repurposing existing drugs. This approach allows AI to quickly identify new therapeutic uses for existing compounds. More than 250 companies are currently using AI in this area. For example, Healx, a UK-based company specialising in rare diseases, used AI to repurpose existing drugs for new therapeutic applications.

In this case, Healx repurposed cimetidine, a drug originally designed to treat stomach ulcers. By leveraging AI, Healx identified cimetidine as a promising treatment for fibrodysplasia ossificans progressiva (FOP), a rare and debilitating genetic disorder.

Although many AI-identified compounds are still under evaluation, the future looks optimistic. As datasets become richer and more accessible, this area is likely to see significant growth.

2. Identifying drug targets:

AI has proven highly effective in building disease models and identifying drug targets or biomarkers more efficiently than traditional techniques. It can process vast amounts of biomedical data, but integrating unstructured datasets continues to pose challenges. AI excels at extracting insights from diverse sources such as academic journals, omics databases, medical imaging, and real-world patient data.

Tools like knowledge graphs have been pivotal in uncovering new relationships between entities. However, their effectiveness depends on the standardisation and labelling of input data. To date, at least 20 drugs with disease-target associations identified through AI are in Phase 1 or 2 trials. By expanding datasets and refining algorithms, companies expect to discover more innovative and entirely novel targets in the coming years.

3. Designing small-molecule drugs:

AI has revolutionised small-molecule drug design by simulating complex chemical properties and generating drug structures with unprecedented speed and accuracy. In this use case, AI tools can either screen chemical libraries or create entirely new compound designs.

A key limitation remains the size and scope of training datasets, which are small compared to the billions of possible compounds in chemical space. Additionally, data availability differs by target class. For example, kinases and G protein-coupled receptors are better characterised, which limits generalisable models and the novelty of resulting drug candidates.

Nonetheless, AI tools for small-molecule drug design have become a cornerstone of the development process. Predictive solutions are continually improving, and AI-designed small molecules outnumber AI-designed antibodies.

Companies such as Exscientia/Recursion and Insilico Medicine are leading efforts in this space, with AI-designed small molecules now in Phase 2 clinical trials. These trials are expected to provide valuable insights into the maturity of AI technologies in drug design.

4. Designing antibody drugs:

AI is gaining momentum in antibody drug design, focusing on optimising current structures and creating new candidates. Compared to small molecules, designing antibodies is more complex due to the computational demands of working with larger molecules.

One major obstacle is the lack of sufficient datasets for antibody sequences and interactions. A significant portion of training data is derived from conventional antibody design libraries, perpetuating challenges like balancing specificity and affinity.

Despite these difficulties, interest in AI-driven antibody design is growing. Over the past year, numerous pharmaceutical companies have disclosed partnerships with start-ups or introduced internal AI capabilities. Notably, Xaira Therapeutics secured over $1bn in funding to focus on de novo antibody design, employing experts in generative modelling and genomics.

This influx of resources and expertise is expected to drive the maturity of AI applications in antibody drug design, enhancing the potential for novel therapeutic candidates.