Spotlight commentary: Integrating artificial intelligence in clinical pharmacology: Opportunities, challenges and ethical imperatives

摘要

The integration of artificial intelligence (AI) into clinical pharmacology heralds a transformative era in healthcare, promising greater efficiencies in drug discovery, personalized medicine and patient care. AI, defined as the simulation of human intelligence by machines, encompasses various subsets including machine learning (ML) and deep learning (DL). ML involves algorithms that enable computers to learn from and make predictions based on data, while DL, a subset of ML, utilizes neural networks with many layers to analyse complex datasets. AI's capacity to process vast datasets exceeds that of the conventional methodologies, providing new insights into drug efficacy and patient outcomes. These large datasets are crucial for work with AI algorithms and include types of data analysis particularly relevant for pharmacology such as natural language processing (NLP), large language models (LLMs), voice recognition, numerical data, multi-omic data integration and image processing. This spotlight commentary explores the multifaceted roles of AI in clinical pharmacology (Figure 1), addressing its potential in accelerating drug discovery, optimizing clinical trials and navigating the ethical landscape associated with its adoption by highlighting a few relevant recent publications on the topic. We start the overview with the paper by Saikin et al. (2019)1 who advocated integration of AI and ML tools into a unified, closed-loop discovery platform. This approach may be crucial for transforming early-stage drug development, where the ability to rapidly test hypotheses and iterate findings can significantly shorten the time to discovery. Closed-loop systems utilize AI not only to design experiments and predict outcomes but also to automatically adjust the experimental parameters based on real-time data. This results in a highly efficient cycle of hypothesis generation, testing and learning which is much faster than traditional drug discovery methods. By integrating diverse AI tools—from predictive analytics to robotic process automation—these platforms could handle complex datasets and experimental conditions, improving the accuracy and speed of identifying viable drug candidates. Furthermore, Lou and Wu (2021)2 focused on the application of AI in identifying new drug-target pairs, particularly for diseases with well-understood mechanisms and drugs exhibiting medium chemical novelty. The use of AI in this context would allow for the exploration of vast chemical spaces and biological interactions more comprehensively and rapidly than human capabilities allow. AI algorithms analyse existing data to predict how new compounds would interact with specific targets, which is crucial for the development of targeted therapies. By leveraging pattern recognition and ML, AI tools can uncover nonobvious relationships between compounds and biological targets, thereby opening up new avenues for therapy that might not have been discovered through conventional research methods. This approach could be very helpful in drug repurposing, where AI identifies novel therapeutic uses for existing drugs, potentially accelerating the drug development process and reducing costs. While the repurposing of antiviral and anti-inflammatory drugs for COVID-19 did not yield successful outcomes despite extensive data and AI efforts, recent advancements in AI have significantly impacted drug discovery and development. Zakeri (2024)3 highlighted how AI-assisted experiments in adeno-associated virus gene therapy research have identified novel pathways and mechanisms, paving the way for new therapeutic approaches. Similarly, Verma et al. (2024)4 demonstrated the potential of AI in drug repositioning, where existing drugs are repurposed for new therapeutic uses by uncovering novel targets and mechanisms of action. Satpathy (2024)5 further emphasized the role of advanced AI algorithms in drug design and development, showcasing their effectiveness in classifying and screening of