(198a) Evaluation on a Combined Cocrystal Screening Method and Its Application on Synthesis of Multicomponent Crystals

Multicomponent crystals, especially the salts and cocrystals, are increasingly being recognized for their potential to modify the physicochemical properties of molecules while maintaining their molecular structures. However, successful synthesis of cocrystals often involves numerous experimental trials, which can be both time-consuming and expensive. To address this challenge, various theoretical predictive methods have been explored to identify potential coformers for cocrystal formation. These predictive methods aim to guide experimental screening and reduce the number of trials and developmental costs associated with cocrystal synthesis.

One such method is the COSMO-RS method, which is an interaction-based approach that relies on the statistical physics of interacting molecular surface segments. This method assesses the miscibility of two components in a supercooled liquid phase based on their excess enthalpy (ΔH_ex), which is the difference between the enthalpy of mixture and those of the pure components. The more negative the ΔH_ex is, the more favorable the two compounds are cocrystallization.Molecular complementarity (MC) analysis, which is a structure-based method that relies on statistical correlation analysis of cocrystals in the Cambridge Structural Database (CSD). This method suggests that molecules with similar geometrical shapes and polarities tend to cocrystallize together. The Hansen solubility parameter (HSP) method is a thermodynamics-based approach that measures the miscibility of two components in a cocrystallization system. The method is based on the "like dissolves like" theory, which suggests that components with similar miscibility are more likely to form a cocrystal.

Indeed, the ability of coformers to interact with the host compound is influenced by a variety of factors, including thermodynamics, structure, and energy. However, each of the methods mentioned above focuses on a particular aspect related to cocrystallization, and the results from a single screening method cannot guarantee the formation of cocrystals in a given system.

To improve the effectiveness and accuracy of coformer screening, we propose a combination of the aforementioned methods to comprehensively evaluate potential candidates. By taking a comprehensive approach that considers multiple factors, we aim to identify the most promising coformers in the screening process, ultimately leading to a higher success rate in cocrystal formation.

In this study, we investigated the predictive ability of three methods, namely COSMO-RS, MC analysis, and HSP model, in identifying potential coformers for the host compound, 2-Amino-4,6-dimethoxypyrimidine (MOP), with a set of 63 components. MOP is a derivative of pyrimidines and aminopyrimidine, which is biologically important as it can be a component of nucleic acids. The molecule contains both hydrogen bond donor and acceptor groups, i.e., the amino-group and pyrimidine ring, respectively, which make it highly likely to interact with other molecules to form cocrystals or salts, some of which have been previously reported. The molecular structure of MOP is shown in Scheme 1.

To evaluate the predictive ability of the three methods, we compared their results with those obtained from systematic experimental screening. Additionally, we combined the three methods to improve the effectiveness and accuracy of the predictions. Based on the results of the experimental screening, we successfully synthesized 21 new multicomponent crystals, including cocrystals and salts of MOP. We systematically characterized ten of these crystals using single crystal X-ray diffraction analysis, both experimentally and theoretically.

The experimental screening of coformers was performed using liquid-assisted grinding (LAG). The resulting powders were then dried and characterized using powder X-ray diffraction (PXRD) and differential scanning calorimetry (DSC) to identify possible solid forms. Based on LAG results and previously reported multicomponent crystals, we found that forty coformers were able to interact with MOP and form new solid phases.

The results of the coformers screening using predictive tools varied between the different methods. The MC method predicted that 53 coformers would "pass," while the COSMO-RS method predicted 46 coformers would "pass" when the cut-off value was set at -1.00 kcal/mol. Three criteria related to the HSP method were considered, with only the Δδ_t criterion being consistent with experimental results and selecting 24 coformers for the "pass" status. Compared to experimental results, the MC method overestimated cocrystallization yields while the HSP method underestimated them. The COSMO-RS method had the highest hit rate among the three methods.

To assess the reliability of the coformer screening methods, we calculated their success rates. The pass rate (PR) was calculated as the number of coformers predicted as "pass" that yielded new solid phases with MOP in LAG, divided by the total number of coformers predicted as "pass". The fail rate (FR) was calculated as the number of coformers predicted as "fail" that did not yield new solid phases in LAG, divided by the total number of coformers predicted as "fail". The overall rate (OR) was calculated as the number of predicted coformers that matched the experimental results, divided by the total number of predicted coformers. The success rates of the three methods are summarized in Table 1. COSMO-RS had the highest success rate (OR: 84.1%), indicating its reliability in predicting coformers for MOP. Although MC had a lower success rate (OR: 69.8%), it could still provide some direction for experimental screening. HSP had a lower success rate (OR: 46.0%) due to producing more "fail" status than actual results, indicating that the interaction factor played a more important role than the structure factor in the cocrystallization of MOP. HSP's calculation was based solely on the cohesive energy and did not consider shape or geometry factors, which could explain its lower success rate.

To improve the accuracy of the predictions, we attempted to combine the applications of these three methods, as they did not provide the highest accuracy individually (especially MC and HSP).

A coformer meeting the criteria of both methods simultaneously is classified as “pass” status, and the combined method is labeled as A∩B, where A and B refer to two different methods. Therefore, there is MC∩HSP, MC∩COSMO-RS and HSP∩COSMO-RS. Similarly, A∪B refers to that a coformer meets the criterions of A, B or A∩B, namely, MC∪HSP, MC∪COSMO-RS and HSP∪COSMO-RS. However, we found that A∪B included too many coformers with “pass” status, making it unsuitable for accurate cocrystallization predictions. For example, in MC∪COSMO-RS, only three in sixty-three coformers were excluded and sixty coformers were classified as “pass” status. Under this condition, the calculations of PR or FR seemed to be unnecessary. The success rates of the A∩B methods were summarized in Table 2. The combination of MC∩COSMO-RS showed the highest PR (89.7%) and OP (85.7%) success rate among all methods before or after combination. It indeed improved the accuracy and efficiency of the predictions. As for MC∩HSP (OR: 47.6%) and HSP∩COSMO-RS (54.0%), the results were unsatisfactory, likely due to the low accuracy of HSP.

Similar to the A∪B method, A∪B∪C method is unnecessary due to its inclusiveness. Instead, we focused on evaluating the performance of the A∩B∩C method. The success rates of this method are summarized in Table 2. However, it yielded the same success rate as HSP∩COSMO-RS. This was due to the fact that the MC method overestimated the cocrystallization of MOP and did not provide any additional coformers to the HSP∩COSMO-RS method.

In summary, COSMO-RS was the most reliable method for predicting coformers with an overall success rate of 84.1%. MC and HSP had lower success rates of 69.8% and 49.0%, respectively. The combination of MC and COSMO-RS showed the best result with an overall success rate of 85.7%. It was found that the A∪B method was not effective as it included too many coformers with "pass" status, while A∩B method improved the accuracy and efficiency of the predictions.

With the assistance of the predictive results and the experimental outcomes (LAG), we further conducted the synthesis of cocrystals and obtained twenty-one new solid phases. Detailed crystal structure of ten multicomponent crystals (two cocrystals and eight salts) were characterized by SCXRD. The motif formed by the complementary pair of O−H···N/O^–···H−N⁺ and N−H···O^–/N−H···O hydrogen bonds was found in all new crystals, they can further self-assemble to build a larger supermolecule. At the same time, we conducted Hirshfeld surface analysis and molecular electrostatic potential (MEP) surface analysis using quantum chemistry theory. Our findings showed that the formation of salts or cocrystals with MOP was dependent on the maximum MEP value of the coformer's carboxyl group, with a higher MEP value indicating a greater likelihood of forming salts.

This work not only further enriched the MOP solid forms but also evaluated three different coformer screening method as well as its combinations. The results showed that the COSMO-RS method had excellent performance in predicting the cocrystallization of MOP. When combined with the MC method, which incorporated the structural factor into the prediction, it yielded the best result, with an overall success rate of 85.7% (PR: 89.7%, FR: 79.2%).