Solid-Form Transition Temperature Prediction from a Virtual Polymorph Screening: A Reality Check

Abstract

The focus of this study is an estimation of uncertainty of solid-form transition temperature (T tr) prediction based on modern virtual polymorph screening calculations. That was done through error propagation, utilizing estimated uncertainties of the relative free energy predictions at 0 K as well as of finite-temperature contribution to the polymorphic relative free energy. It was found that the uncertainty of the T tr prediction displays an inverse dependence on the difference of slopes of the intersecting relative free energy curvesthe lower the difference, the higher the uncertainty of T tr prediction. The results demonstrated that the error of T tr prediction is expected to be high (up to ∼260–270 K), especially at the temperature range close to ambient and above. The lowest uncertainty of T tr prediction at room temperature is expected to be ∼50 K in the case of virtual forms being separated by 10 kJ/mol at 0 K.

Prediction of the Relative Free Energies of Drug Polymorphs above Zero Kelvin

“…Using different DFT rankings of the lattice energies at zero Kelvin (see Figures , S2, and S3), it can be seen that this can impart a significant change in the transition temperatures predicted by the free energy profiles. Consequently, it is very challenging to accurately predict transition temperatures due to the uncertainty introduced from both the QM and MD calculations. However, the trend of change and relative ranking of stability with temperature is qualitatively correct.…”

“…For example, in the present study using optPBE-vdW the rankings were found to be B < C < E < A < D. Also, in the sixth blind test Form B was predicted to be the lowest energy Form by the majority of participants who successfully found more than one experimental form. In a recent detailed analysis of how the level of theory might affect the energetic ordering of the forms, Hoja et al 26 showed that using PBE+TS gave the ordering B < C < D < A, 83 PBE+MBD gave the ordering C < B < E < A < D, and by using a hybrid functional instead of a pure functional (i.e., PBE0+MBD), the same ordering was achieved (C < B < E < A < D). In the present study, we also calculated the lattice energies using B3LYP-D*, which gave the ordering C < B < A < E < D (Figure 5).…”

Virtual Coformer Screening by Crystal Structure Predictions: Crucial Role of Crystallinity in Pharmaceutical Cocrystallization

“…A thermodynamic driving force of cocrystallization is defined by a negative free energy of cocrystal formation, ΔG cc . However, though a finite-temperature contribution to the free energy of a crystalline form could be accurately calculated, , that is a time-consuming procedure which was not considered in this study. Due to the lack of the entropic contributions, the Δ E cci classification cutoff values between positive (a cocrystal is formed) and negative (no cocrystal is found) observations may differ from 0.…”

Uncertainty Distribution of Crystal Structure Prediction

“…The development of a reliable CSP approach faces a number of challenges, many of which have been partially addressed such as dependency on level of theory, , temperature-dependent free energy contributions, − molecular flexibility, − solid form complexity, − kinetic hindrance effects, ,, and acceleration of high-level predictions …”

“…Uncertainty evaluation is important for any tool, experimental or computational, which is applied for practical purposes. However, except for a few studies, ,− attention paid to uncertainty of CSP predictions in the computational organic solid-state field has been limited so far. Since a modern CSP workflow includes iterations of multiple stages (force field fitting, crystal packing generation, clustering, DFT-D ranking at 0 K, and finite-temperature contribution calculations) until convergence is reached, there could be various sources of prediction error.…”