Virtual Multicomponent Crystal Screening: Hydrogen Bonding Revisited

Abstract

Pharmaceutical cocrystals, salts, and multicomponent crystals, in general, have increasingly come under the spotlight in recent years. A fast and efficient a priori theoretical classifier to identify potential coformers is highly sought after to complement the experimental brute force screening methods. This research examines the qualitative approaches that are based on hydrogen bonding strength. First, molecular electrostatic potential (MEP) maps of 330 coformers were obtained from density functional theory simulations, using two geometries: experimentally determined crystal structures and gas-phase optimization. An in-depth comparison of MEPs revealed the potential pitfalls of these two geometries that are deliberated at length in the manuscript. Next, six APIs and their reported salts/cocrystals on the Cambridge Structural Database (CSD) were inversely predicted with MEP analysis. For two of these APIs, the prediction showed systematic errors that are resolved with suggestions provided in the manuscript. Subsequently, hydrogen bond energy (HBE) and hydrogen bond propensity (HBP) calculations were put to the test with two APIs and 52 organic coformers. Finally, multivariate logistic regression, a linear machine learning (ML) algorithm, showed how a combination of HBE and HBP can be a superior classifier, for which 18 out of 25 positive cases were uninterruptedly identified at the top of the list. Provided that a database of failed attempts of cocrystallization is compiled within the scientific community to supplement the existing positive results (multicomponent crystals in the CSD), the combination of chemistry-based parameters and ML can be a promising classifier for coformer selection.

A study to discover novel pharmaceutical cocrystals of pelubiprofen with a machine learning approach compared

“…Particularly, compared to other virtual screening methods such as MEPS and HBP, we believe that our approach is competitive in the absence of complex quantum This journal is © The Royal Society of Chemistry 2022 calculation, easy to access (free of charge), and saving computational time. 46 From the point of view of outcomes, one may argue that the ANN model is not quite effective because only two, INA and NCA, turned out to form cocrystals with PF among the 27 candidate coformers that have probabilities greater than 0.5. However, this work was based on a limited number of compounds, and screening with a more extensive list of coformers may change these figures.…”

“…Particularly, compared to other virtual screening methods such as MEPS and HBP, we believe that our approach is competitive in the absence of complex quantum calculation, easy to access (free of charge), and saving computational time. 46…”

“…MEPS analysis is an efficient tool for the clarification of various chemical properties such as chemical reactivity, intermolecular interactions, and charge distribution. 46 Cocrystal formation is profoundly influenced by the hydrogen bond donor (positive potential) and acceptor (negative potential) propensities of the cocrystallizing compounds; thus, the charge distribution obtained from MEPS analysis can be used to predict the molecular recognition patterns in the cocrystals. through the stable homosynthon (synthon B in Fig.…”

Unravelling the Origin of Solvate Formation in the Anticancer Drug Trametinib: Insights from Crystal Structure Analysis and Computational Modeling

“…31 Polar protons (attached to nitrogen or oxygen) were modeled using the potential parameters from Coombes et al, 32 and S atoms were modeled using the potential parameters used by Price et al in the first blind test for crystal structure prediction. 33 These sets of intermolecular potential parameters were supplemented with the following set of I•••I potential parameters: 34 A ik = 13968.1 eV, 1/B ik = 0.35 Å, C ik = 476.4 eV Å 6 . Duplicate lattice energy minima following the DMACRYS optimizations were removed using customized python scripts that make use of…”

“…In this regard, the concepts of hydrogen bond propensity (HBP) and hydrogen bond energy (HBE) has found relevance in being able to predict the outcome of cocrystallization events. 6 Furthermore, approaches based on machine-learning (ML) 7 and molecular dynamics simulations 8 directed toward the ability to predict cocrystal formation with a large number of coformers have also been performed. In addition, crystallization events can result in the formation of polymorphs, solvates, as well as the occurrence of high Z′ structures of the active pharmaceutical ingredient (API).…”