Machine learning study of the molecular drivers of natural product prices

Abstract

AbstractThe price of chemicals is a very complex variable. It can be impacted by production costs but also by market and managerial factors, which may have complex relationships with molecular characteristics and the state of technology and society. In this work, we explore the extent to which molecular characteristics can help explain natural product prices with the aid of machine learning tools. We interpret models trained on molecular descriptors and molecular fingerprints. These models can explain a notable proportion of the variation in prices, suggesting that production and separation costs are a major contributor to current natural product prices. Some molecular properties stand out as key price drivers across the chemical space, including hydrophobicity and the presence of certain heteroatoms. On the other hand, we demonstrate how the application of cliff analysis to prices allows the identification of small chemical transformations that have a remarkable impact on prices. Overall, the work suggests that machine learning could help achieve more consistent and fairer pricing and provides specific examples of chemical transformations in which synthetic biology could add significant value. © 2021 Society of Chemical Industry and John Wiley & Sons, Ltd