1906 Experimental Biology and Medicine Volume 248 November 2023 research. Roberts3 writes a comprehensive review of the challenges and opportunities presented by data science techniques in the context of drug safety assessment. This article highlights the increasing reliance on data-driven approaches in the pharmaceutical industry, emphasizing their potential to expedite drug development, reduce costs, and improve patient safety. It discusses the significance of harnessing vast and diverse datasets, including chemical, biological, and clinical data, to enhance safety prediction and monitoring throughout the drug development lifecycle. This review also explores the challenges faced by data scientists and researchers in the field of drug discovery safety and underscores the need for innovative data integration techniques, scalable computational infrastructure, and standardized protocols to address these challenges effectively. In addition, it explores the myriad opportunities data science affords to pharmaceutical research and development. This article offers an informative resource for researchers, pharmaceutical professionals, and policymakers to foster a deeper understanding of the evolving field of data science in drug discovery safety. The advent of Bidirectional Encoder Representations from Transformers (BERT)-like large language models (LLMs) has revolutionized natural language processing and, in turn, their application in domains like patient safety and pharmacovigilance (PSPV). Wang et al. 4 report an in-depth comparative study on BERT-like LLMs for causal inference in PSPV. Generic pretrained BERT LLMs, domain-specific pretrained LLMs, and domain-specific pretrained LLMs with safety knowledge–specific fine-tuning are compared using three publicly accessible PSPV datasets to assess the influence of data complexity and model architecture, the correlation between the BERT size and its impact, and the role of domain-specific training and fine-tuning. They find that data complexity and model size have little impact on the performance of the BERT-like LLMs, and domain-specific BERT-like LLMs outperform generic pretrained BERT models in causal inference. This study demonstrates the challenges and opportunities associated with utilizing BERT-like models in PSPV, particularly in the context of causal inference. Their findings highlight the importance of BERT-like models for the timely detection of adverse events, signal detection, and the potential for proactive risk management in drug safety, providing insights into the utility of these models in automating the extraction and analysis of valuable information from unstructured textual data sources. In the domain of pharmaceutical research and healthcare, efficient and accurate analysis of drug labeling text is crucial for drug development, safety monitoring, and informed clinical decision-making. Wu et al. 5 introduce RxBERT (a BERT model pretrained on Food and Drug Administration [FDA] human prescription drug labeling documents), a novel approach that leverages advanced natural language modeling techniques to enhance drug labeling text mining and analysis. RxBERT is evaluated with multiple datasets, resulting in slightly better performance than other natural language processing models. RxBERT is a domain-specific language model fine-tuned on a large corpus of drug labeling texts. Their findings demonstrate the significance of RxBERT in advancing drug labeling text mining and analysis in the pharmaceutical and health-care industries.

The opioid crisis in the United States has raised significant concerns regarding the safety and adverse effects of opioid medications …