Implementation and Evaluation of Ensemble Learning Algorithm for Improved Drug Development

Abstract

Artificial intelligence (AI) & machine learning (ML) has emerged as the cutting-edge technologies most expected to revolutionize the pharmaceutical R&D industry during the previous decade (R&D). Part of the reason for this is that barriers that once prevented the collection and processing of massive amounts of data have been largely removed as a result of breakthrough developments in computer technology. Meanwhile, the expense of developing, testing, and delivering a new drug to market and eventually to patients has skyrocketed. The International Health Organization has advocated for pharma covigilance, the monitoring of adverse drug events, as a means of ensuring the security of medicines by facilitating the rapid and trustworthy transmission of information pertaining to drug safety issues. Our goal is to have a conversation about the use of machine learning techniques and causal reasoning paradigms in the field of pharma covigilance. Over the past two decades, ML methods have become increasingly integrated into the pharmaceutical industry's search for new therapeutics. Clinical study design, conduct, & analysis are the most recent areas of drug research to see beneficial disruption from AI/ML. We highlight three current paths or voids being explored to combine causal inference and machine learning in drug safety studies. Finally, our research found that using causal paradigms can help prevent common problems with ML models. Through a series of comparisons, we demonstrate that the proposed ensemble algorithms outperform state-of-the-art ML prediction algorithms over a wide range of metrics.