AI in chemistry: accelerating early-stage drug discovery through the iterative use of machine learning and improved data quality
Date
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Currently, it takes around 2 billion dollars and 15 years to bring a drug to market. The excessive cost and time can be partly attributed to issues within the drug discovery pipeline. This research focuses on speeding up the time it takes to bring a drug to market as well as the associated cost. This is done by applying machine learning methods to early-stage drug discovery to improve hit rates, improve the quality of data, and investigate the size of the dataset needed for training ML models. More specifically, this study investigates the predictive power of ML models in the classification of growth inhibitory activity, colloidal aggregation, and enzyme inhibition. Furthermore, in the case of colloidal aggregation and enzyme inhibition, a subset of ML known as xAI (explainable AI) is utilized to gain insights into how the model makes predictions by having said model explain itself. While it is not yet clear whether this work has directly accelerated the drug discovery pipeline, it has contributed meaningfully to the process. Notable outcomes include: the development of a ML model used in identifying a lead compound, a predictive xAI model for small molecules for identifying and modifying aggregation, and a foundation for an xAI model that predicts enzyme inhibition of small molecules. Additionally, this work has improved data quality and provided insights into the optimal size and balance of training datasets for ML models.