Use of Artificial Intelligence in Drug Development

This cross-sectional study investigates the scope and breadth of artificial intelligence use in drug development.


Introduction
Considerable focus has been placed on the health care applications of artificial intelligence (AI).
Already, machine learning, a subset of AI that involves "the use of data and algorithms to imitate the way that humans learn" 1 has been used to predict diseases, 2 while AI-powered smartphone apps have been developed to promote mental health and weight loss. 3Owing in part to such successes, the market for AI in health care has been forecasted to increase more than 1000% between 2022 and 2029, from $13.8 billion to $164.1 billion. 4e area of substantial promise is drug development, which is poised to benefit from advances in the use of AI to predict protein folding, molecular interactions, and cellular disease processes. 5ccessful application of AI to drug development, however, faces several obstacles, including poor model performance caused by nondiverse training data and shortcut learning.Additionally, the often opaque ways that AI systems reach their predictions conflict with regulatory approval frameworks that require a rationale for decision-making.Given these obstacles, we sought to identify the scope and breadth of AI use in drug development.

Methods
We conducted a cross-sectional study of investigational and approved drugs (n = 102 454) listed in the global research and drug development database Pharmaprojects (Informa, Citeline) on February 11, 2024.Institutional review board approval was not required because this study did not involve human participants, in accordance with 45 CFR §46.We followed the STROBE reporting guideline.
To identify AI-developed drugs, we used AI search terms from Janiesch et al 6 and the National Library of Medicine's Medical Subject Headings database (eAppendix in Supplement 1).Automated scans of information for each drug were evaluated.If a drug was described as developed with AI, the type and purpose of AI use were noted.When the type or purpose of AI use could not be determined, additional information was obtained from internet sources.The most specific term identified was used to categorize the AI application (Figure 1).Among AI-developed drugs, descriptive statistics were compiled of AI type, AI purpose, therapeutic area, and development status.Statistical analysis was performed using Excel version 16 (Microsoft) from February to March 2024.

Results
The database search yielded 406 drugs, of which 241 were excluded upon review for no reported use of AI in drug development.AI use was reported in the development of 164 investigational drugs and   Role of the Funder/Sponsor: The funders had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.

1
approved drug.The most frequent types of AI use were machine learning (n = 46 [28%]) and deep learning (n = 28 [17%]).AI was used for 12 purposes, most commonly drug molecule discovery (n = 125 [76%]) (Figure 2).Examples of such use ranged from platform screening of drugs, in which AI was used to analyze molecular images of the effects of drugs on a cell, to deep generative modeling to design virtual novel molecules.Modest AI use was observed for drug target discovery (n = 37 [22%]), including machine learning to find previously unknown connections between genomic, chemical, and clinical data.AI use for clinical outcomes analysis, such as the use of AI-based+ Supplemental contentAuthor affiliations and article information are listed at the end of this article.