PS27: Harnessing AI/ML to Transform Pharmaceutical Development: Scientific Accomplishments of the AIML Scientific Working Group Initiative

The Statistical Perspectives of AI/ML Scientific Working Group (SWG) is dedicated to exploring, developing, and disseminating best practices for applying artificial intelligence (AI) and machine learning (ML) across pharmaceutical research and development. This initiative addresses the urgent need for a structured, multidisciplinary approach to responsibly and effectively integrate AI/ML techniques in drug discovery, clinical development, and related domains. Through collaboration among statisticians, clinicians, AI researchers, and regulatory experts, the SWG aims to drive innovation while upholding ethical standards and regulatory compliance. The SWG's key objectives include advancing AI/ML methodologies for drug discovery, optimizing clinical trial design and execution, leveraging real-world data to generate evidence for regulatory decisions, and enhancing drug safety monitoring and pharmacovigilance. The group also prioritizes statistical perspectives to address algorithmic bias, protect patient privacy, and ensure transparency in AI-driven processes. This session will showcase research and accomplishments from SWG sub-streams focusing on preclinical research, clinical development, medical affairs, and pharmacovigilance. These sub-streams operate under goals defined by a steering committee of cross-disciplinary experts. By aligning stakeholders under a unified vision, the SWG promotes the responsible and transformative application of AI in pharmaceutical development, ensuring regulatory alignment and patient-centered outcomes. Aligned with the ASA Biopharmaceutical Section's mission, this session underscores the pivotal role of biostatisticians in advancing AI/ML in pharmaceutical research, highlighting significant progress achieved in the SWG's first year. Organizer: Margaret Gamalo Speakers: 1. Ixavier Higgins/Yushi Liu: Preclinical Subteam Session: Connecting the Dots: Leveraging AI/ML Tools to Increase Interpretability of Genomic Signatures from Mouse to Human Mouse models have long been used to characterize human diseases and drug signatures. However, translating genomic changes between species can be quite challenging due to biological differences. This disconnection contributes to the very low success rate of drug translation from preclinical studies to clinical trials. Therefore, understanding the linkages between human and mouse remains crucial. In our study, we utilized PBMC mouse and human single-cell RNA-seq time course data under trauma conditions to conduct transfer learning analysis. We carefully examine the characteristics of genes with good translatability and explore their linkage to 'causal' genes. Collectively, our work aims to enhance the understanding of disease mechanisms and drug actions based on animal models. 2. Li Wang/ Yue Shentu: The Clinical Development sub-team conducted an industry-wide survey on AI/ML use cases in clinical development. The summary, results and insights of this survey will be shared in the session. The Clinical Development effort is organized into three sub-streams: 1. Precision Medicine 2. Prognostic Modeling 3. Other Topics: This includes imaging endpoints, generative AI using Twin methodology, inclusion/exclusion criteria (IC/EC) optimization, and LLM automation. Each sub-stream has initiated comprehensive literature reviews focusing on both theoretical perspectives and practical applications in these areas. Summaries and use cases from these reviews will be presented in the session to provide insights and foster knowledge sharing. 3. Achilleas Livieratos/Junjing Lin: Advancing Clinical Evidence Generation through LLMs: Use Cases in Network Meta-Analysis and Pharmacovigilance through the Sentinel Initiative The integration of artificial intelligence (AI) and advanced statistical frameworks is transforming evidence-based medicine, particularly in network meta-analysis (NMA) and pharmacovigilance. This presentation explores two groundbreaking use cases: (1) automating NMAs using AI-powered frameworks and (2) enhancing drug safety monitoring through the FDA Sentinel Initiative. The NMA use case leverages large language models (LLMs) and the Mixture-of-Agents (MoA) framework to automate clinical data extraction, Bayesian modeling, and treatment comparisons. Using open-source tools like PyMC and GPT-4, our approach streamlines clinical evidence synthesis while ensuring scalability, accuracy, and transparency through rigorous evaluation metrics such as FLASK (Fidelity, Logic, Accuracy, Succinctness, Knowledge). The second use case highlights the FDA Sentinel Initiative's integration of real-world data (RWD) and AI-driven natural language processing (NLP) for adverse drug reaction (ADR) detection. By combining structured claims data with unstructured electronic health records (EHRs), the Sentinel system identifies critical safety signals and improves real-time pharmacovigilance. Together, these use cases demonstrate how AI-driven frameworks can enhance clinical research, regulatory submissions, and drug safety monitoring, offering scalable, data-driven solutions that optimize healthcare outcomes through timely and precise medical evidence generation.