Artificial Intelligence on Drugs: A Large-Scale Examination of AI and Drug Innovation
55 Pages Posted: 20 Feb 2020 Last revised: 17 Jun 2020
Date Written: June 15, 2020
As a potential general-purpose technology, AI has been touted to accelerate drug discovery. Yet, despite substantial investment, drug development has slowed down in terms of finding new drugs with new chemical properties. To understand how AI affects drug discovery, we examine and differentiate its effects on different stages of drug development. Using patents and job postings to measure AI, we find that AI can primarily affect the earliest stage on compound discovery when tasks are heavily dependent on automatic data processing and reasoning to identify drug-target pairs. However, AI is less useful in later stages when human engagements, judgements, and organizational decisions are essential. Furthermore, despite discovering more drug candidates at the early stage, AI has not discovered higher quality drugs that are more likely to get approved than others. Thus, with the bottleneck of drug development remaining in the later stages, AI has not changed the overall number of drugs that clear final approval. To further examine the conditions on when AI is most effective in aiding drug discovery, we find AI has an advantage in discovering drugs whose mechanism of impact on a disease is known and drugs at the medium level of chemical novelty. AI is less helpful in discovering drugs when there is no existing therapy or drugs at either extreme end of the novelty spectrum—those that are either entirely novel or those that are incremental “me-too” drugs. Taken together, our study sheds light on both the advantages and the limitations of using AI in drug development.
Keywords: AI, Drug Discovery, Machine Learning, Economics of AI
Suggested Citation: Suggested Citation