De-Risking Early-Stage Drug Development With a Bespoke Lattice Energy Predictive Model: A Materials Science Informatics Approach to Address Challenges Associated With a Diverse Chemical Space

Abstract

The solid-state properties of new chemical entities are critical to the stability and bioavailability of pharmaceutical drug products. The stability of the solid-state packing is described by the packing energy and an accurate prediction of this property for drug molecules would therefore be desirable. However, this has been difficult to achieve because of the lack of fundamental thermodynamic data on drug molecules. A potential solution would be to use calculated lattice energies to build a model and design molecules with desired physicochemical properties from an early stage, aligning with a “design by first intent” strategy for physicochemical properties. We first demonstrate the high correlation and interchangeability between QSPR models built using calculated lattice energies and experimental sublimation enthalpies for small organic molecules. We then present a QSPR model trained on in-house molecules using lattice energies calculated from crystal structures. The result is a model that enables fast prediction of the lattice energies of in-house molecules from 2-D molecular structure with reasonable accuracy (R2 = 0.92, root mean square error = 3.58 kcal/mol). We explore the model elements to improve our understanding of the molecular properties that contribute to lattice energy and then suggest potential cross-industry aspects that may enhance the application of the concept.