AI in drug design: evolution or revolution?
Darren sets out to explore the impact of artificial intelligence (AI) on the pharmaceutical industry, particularly in the drug discovery sector. He begins by setting out the different AI methods, such as machine learning (ML; models that learn from previous data), deep learning (ML models that mimic human brain processes), generative AI (creating original data), to the more traditional and evolution of quantitative structure-activity relationship (QSAR) models. Green then states that a lot of these methods are not new, but with the rise of computational power and techniques, they have produced an acceleration in methods. Several of which are outlined in the review, including but not limited to: large-scale cheminformatics (such as matched molecular pairs and series); large enumeration of readily accessible chemical space, free energy perturbation (FEP, invented in 1987 and now usable); deep learning QSAR models; ML-based forcefields with DFT levels of accuracy and accurate protein structure prediction (Alphafold and Rosetta Fold). Additionally, there is discussion about several individual successful applications of ML in small molecule and biologics discovery, most notably the SARS-CoV-2 PLpro inhibitor. They then go on to discuss whether AI is there yet and the current limitations of needing high-quality scientific data to create good models, which is expensive, and that current methods are poor at extrapolating with a limited understanding of what the models are learning. The final comments are that AI will be continued to be used within the drug discovery process and greater developments will be continued to be made and companies will either be built around AI or will have to remodel to embrace AI. We are just at the beginning of the AI revolution.