A Survey of AI for Drug R&D and Why Explainable AI Matters

Khai Pham, ThinkingNodeLife.ai ABSTRACT: To bring a new drug to the market, the average cost is around $2 billion, and it takes approximately 10 to 12 years. This trend has become more challenging as the development of “low hanging fruit” drugs has already taken place. However, the advancements in computing hardware and software, along with the adoption of AI/ML, have shown promising results by shortening the time from discovery to clinical trials. The successful application of AI in drug R&D depends on the availability of high-quality data and knowledge and the use of the right type of AI based on the development phase. Contrary to popular belief, drug R&D is primarily knowledge-driven and not solely data-driven. For instance, applications like self-driving cars are data-driven because they mainly require statistical pattern recognition capabilities without the need for in-depth comprehension of the decision processes. In contrast, drug R&D requires explainable AI, which is not only vital to ensure transparency, accountability, and trustworthiness of the models but also to understand the pathobiology of the disease and the mechanisms of action of the drug for better intervention strategies to increase the efficacy and lower the risk of negative side effects. The successful development of efficient explainable AI requires the understanding of the level of explanation that is needed for a given audience and is based on the type of AI that has been used.