Precision Pharmacology and Pharmacovigilance: Leveraging AI to address drug safety knowledge gaps

Cedars-Sinai Medical Center, West Hollywood CA

Investigators

Linked publications, trials & patents

Abstract

I am proposing to continue my research program in precision pharmacology and pharmacovigilance coupling observational data analysis with recent advancements in artificial intelligence to advance drug safety and efficacy. Over the past five years, the availability of observational data via electronic health records and biobanks, both institutional and national, has grown dramatically. It is not difficult to imagine a world where our doctors will have the evidence and knowledge to provide personalized guidance and treatment to maximize health and longevity. And that, in turn, the data generated by these encounters be collected, organized, and analyzed by biomedical researchers to invent the next generation of interventions. This R35 project has produced 19 papers covering topics ranging from digital health and machine learning in medicine and epidemiology and drug safety studies to studying the molecular underpinnings of adverse drug reactions. However, significant challenges remain that prohibit meaningful progress. I have identified three major challenges that I plan to address over the next five years: There remains high variability in adverse drug events, with some patients experiencing significant burden, while others are relatively symptom free; The complex multimodal structure and non-random missingness of observational clinical data make data analysis difficult; Limited understanding of the molecular mechanisms of drug reactions and drug-drug interactions challenges both prediction and validation. No one solution can address all these challenges. They each necessitate a distinct blend of data science, informatics, and experimental approaches. These challenges are extensions of the previous R35 challenges and refined through our experience these past five years. For example, we have significantly advanced pediatric drug safety by developing novel informatics models to identify safety signals across child development. We demonstrated the importance of integrating clinical trial data and drug safety databases with dynamic biological mechanisms, aiming to systematically understand age-dependent drug toxicities and genetic susceptibilities. While the major challenge areas are distinct, they complement each other. Investigating the variability in adverse drug event severity and frequency (Challenge #1) will independently lead to better treatment decisions for the affected patients. In addition, this work will be complemented by our work to integrate observational data (Challenge #2) by generating new drug safety hypotheses and integrating molecular data (Challenge #3) will guide follow up experiments for validation and further discovery. The above-mentioned challenges are substantial and may not be resolved within the next five years. However, tackling these issues promises to yield groundbreaking insights, enhancing drug design, furthering precision medicine, and shaping the future of drug safety governance.