It takes on average at least 10 years to get a new drug to the market. The drug discovery and development process is long and expensive. The process comprises a chain of complex procedures, including early stage (target identification and validation, hit discovery and lead optimisation), preclinical trials, clinical trials and finally review and post-approval research.
Many in the field commonly describe the process of finding a new drug as finding a needle in a haystack. Nowadays, computer-aided drug discovery is a well-used tool in the drug discovery process. However, there is an urgent need for new approaches that can improve and optimise the drug discovery and development pipeline. Recently, AI technologies, particularly machine learning and deep learning, have been incorporated into this pipeline in a variety of ways. The enormous potential of AI in drug discovery can be highlighted by the growing relationship between AI companies and the pharma industry.
AI in drug discovery
Here are just some of the key applications of AI in early drug discovery:
- Target identification and validation – With increasing amounts of publicly available biological data, AI is an ideal candidate to identify molecular targets. For example, a recent study used a Bayesian machine learning approach to predict drug binding targets.
- Hit and lead identification –
- Virtual screening (VS) – A series of computational techniques that aim to prioritise a large chemical library. The application of AI could enhance VS efficiency and accuracy.
- Drug repurposing – An appealing alternative for drug discovery is to find a new use for an old drug. Machine learning approaches can be both direct (i.e. classifying therapeutic categories) or indirect (i.e. identifying potential lead compounds from existing compounds).
- Generative models and de novo design – Generative molecular designs based on deep learning have garnered a lot of attention in recent years. AI in de novo design plays two fundamental roles. First, it provides an algorithm to effectively generate molecules. Second, it evaluates the generated compounds.
- Property prediction – Once scientists have identified a series of compounds they then need to be optimised in terms of the properties that regulate their behaviour within an organism. The pharmacokinetic profile, such as their ADME properties, and their toxicological effects need to be improved. AI can drive this process, for example, a recent study used a deep learning methodology to predict ADMET properties.
The ‘hype versus reality’ topic is a recurrent discussion related to the implementation of AI in drug discovery. While expectations are high, there have been several notable advances. It is not yet clear how far we are from a new era of AI-driven drug discovery. However, it is apparent that these technologies are here to stay. In the coming years, experts expect the number of AI applications to increase. A key aspect of the successful implementation of AI will be consideration of the interpretability of the models. While this is a hard task, the authors noted that cooperation between chemists and data scientist will be an important step towards more easily interpretable models.