The numbers that have long defined traditional drug discovery are, by any measure, daunting. Bringing a single
new medicine to market can take over a decade and costs on average over $1billion. Despite this investment, only around 10% of drug candidates that enter clinical trials ever receive regulatory approval, and a smaller fraction of drug discovery programmes result in a marketable therapy.

These figures reflect not only economic inefficiency but also the limitations of existing discovery approaches in translating biological insight into effective therapeutics.

Conventional approaches rely heavily on high-throughput screening, which involves testing vast libraries of compounds against a target and hoping something binds. While effective for a subset of well-characterised and structurally accessible targets, this model is poorly suited to proteins that are dynamic, membrane-bound or insufficiently characterised. Up until recently, these have been referred to as “undruggable” targets, not impossible to drug in theory, but resistant to every conventional approach available in practice.

Recent advances in artificial intelligence (AI), particularly generative modelling, are beginning to address these limitations. Rather than screening existing chemical or antibody libraries, AI-enabled approaches allow for the design of novel molecules with properties optimised for specific biological targets.

This article will explore why traditional drug discovery has fallen short, what the emergence of AI-driven drug discovery has done to change the game and what the future holds for this new technology.