General Graph Neural Network-Based Model To Accurately Predict Cocrystal Density and Insight from Data Quality and Feature Representation

Cocrystal engineering as an effective way to modify solid-state properties has inspired great interest from diverse material fields while cocrystal density is an important property closely correlated with the material function. In order to accurately predict the cocrystal density, we develop a graph neural network (GNN)-based deep learning framework by considering three key factors of machine learning (data quality, feature presentation, and model architecture). The result shows that different stoichiometric ratios of molecules in cocrystals can significantly influence the prediction performances, highlighting the importance of data quality. In addition, the feature complementary is not suitable for augmenting the molecular graph representation in the cocrystal density prediction, suggesting that the complementary strategy needs to consider whether extra features can sufficiently supplement the lacked information in the original representation. Based on these results, 4144 cocrystals with 1:1 stoichiometry ratio are selected as the dataset, supplemented by the data augmentation of exchanging a pair of coformers. The molecular graph is determined to learn feature representation to train the GNN-based model. Global attention is introduced to further optimize the feature space and identify important atoms to realize the interpretability of the model. Benefited from the advantages, our model significantly outperforms three competitive models and exhibits high prediction accuracy for unseen cocrystals, showcasing its robustness and generality. Overall, our work not only provides a general cocrystal density prediction tool for experimental investigations but also provides useful guidelines for the machine learning application. All source codes are freely available at https://github.com/Xiao-Gua00/CCPGraph.

FDA/M-CERSI Co-Processed API Workshop Proceedings

These proceedings contain presentation summaries and discussion highlights from the University of Maryland Center of Excellence in Regulatory Science and Innovation (M-CERSI) Workshop on Co-processed API, held on July 13 and 14, 2022. This workshop examined recent advances in the use of co-processed active pharmaceutical ingredients as a technology to improve drug substance physicochemical properties and drug product manufacturing process robustness, and explored proposals for enabling commercialization of these transformative technologies. Regulatory considerations were discussed with a focus on the classification, CMC strategies, and CMC documentation supporting the use of this class of materials from clinical studies through commercialization. The workshop format was split between presentations from industry, academia and the FDA, followed by breakout sessions structured to facilitate discussion. Given co-processed API is a relatively new concept, the authors felt it prudent to compile these proceedings to gain further visibility to topics discussed and perspectives raised during the workshop, particularly during breakout discussions. Disclaimer: This paper reflects discussions that occurred among stakeholder groups, including FDA, on various topics. The topics covered in the paper, including recommendations, therefore, are intended to capture key discussion points. The paper should not be interpreted to reflect alignment on the different topics by the participants, and the recommendations provided should not be used in lieu of FDA published guidance or direct conversations with the Agency about a specific development program. This paper should not be construed to represent FDA's views or policies.

Cocrystal Prediction of Bexarotene by Graph Convolution Network and Bioavailability Improvement

Bexarotene (BEX) was approved by the FDA in 1999 for the treatment of cutaneous T-cell lymphoma (CTCL). The poor aqueous solubility causes the low bioavailability of the drug and thereby limits the clinical application. In this study, we developed a GCN-based deep learning model (CocrystalGCN) for in-silico screening of the cocrystals of BEX. The results show that our model obtained high performance relative to baseline models. The top 30 of 109 coformer candidates were scored by CocrystalGCN and then validated experimentally. Finally, cocrystals of BEX-pyrazine, BEX-2,5-dimethylpyrazine, BEX-methyl isonicotinate, and BEX-ethyl isonicotinate were successfully obtained. The crystal structures were determined by single-crystal X-ray diffraction. Powder X-ray diffraction, differential scanning calorimetry, and thermogravimetric analysis were utilized to characterize these multi-component forms. All cocrystals present superior solubility and dissolution over the parent drug. The pharmacokinetic studies show that the plasma exposures (AUC0-8h) of BEX-pyrazine and BEX-2,5-dimethylpyrazine are 1.7 and 1.8 times that of the commercially available BEX powder, respectively. This work sets a good example for integrating virtual prediction and experimental screening to discover the new cocrystals of water-insoluble drugs.

Applications of Artificial Intelligence and Machine Learning Algorithms to Crystallization

Artificial intelligence and specifically machine learning applications are nowadays used in a variety of scientific applications and cutting-edge technologies, where they have a transformative impact. Such an assembly of statistical and linear algebra methods making use of large data sets is becoming more and more integrated into chemistry and crystallization research workflows. This review aims to present, for the first time, a holistic overview of machine learning and cheminformatics applications as a novel, powerful means to accelerate the discovery of new crystal structures, predict key properties of organic crystalline materials, simulate, understand, and control the dynamics of complex crystallization process systems, as well as contribute to high throughput automation of chemical process development involving crystalline materials. We critically review the advances in these new, rapidly emerging research areas, raising awareness in issues such as the bridging of machine learning models with first-principles mechanistic models, data set size, structure, and quality, as well as the selection of appropriate descriptors. At the same time, we propose future research at the interface of applied mathematics, chemistry, and crystallography. Overall, this review aims to increase the adoption of such methods and tools by chemists and scientists across industry and academia.

Prediction of Mefenamic Acid Crystal Shape by Random Forest Classification

OBJECTIVE

Particle shape can have a significant impact on the bulk properties of materials. This study describes the development and application of machine-learning models to predict the crystal shape of mefenamic acid recrystallized from organic solvents.

METHODS

Crystals were grown in 30 different solvents to establish a dataset comprising solvent molecular descriptors, process conditions and crystal shape. Random forest classification models were trained on this data and assessed for prediction accuracy.

RESULTS

The highest prediction accuracy of crystal shape was 93.5% assessed by fourfold cross-validation. When solvents were sequentially excluded from the training data, 32 out of 84 models predicted the shape of mefenamic acid crystals for the excluded solvent with 100% accuracy and a further 21 models had prediction accuracies from 50-100%. Reducing the feature set to only solvent physical property descriptors and supersaturations resulted in higher overall prediction accuracies than the models trained using all available or another selected subset of molecular descriptors. For the 8 solvents on which the models performed poorly (< 50% accuracy), further characterisation of crystals grown in these solvents resulted in the discovery of a new mefenamic acid solvate whereas all other crystals were the previously known form I.

CONCLUSIONS

Random forest classification models using solvent physical property descriptors can reliably predict crystal morphologies for mefenamic acid crystals grown in 20 out of the 28 solvents included in this work. Poor prediction accuracies for the remaining 8 solvents indicate that further factors will be required in the feature set to provide a more generalized predictive morphology model.

Progress in Research on Artificial Intelligence Applied to Polymorphism and Cocrystal Prediction

Artificial intelligence (AI) is a technology that builds an artificial system with certain intelligence and uses computer software and hardware to simulate intelligent human behavior. When combined with drug research and development, AI can considerably shorten this cycle, improve research efficiency, and minimize costs. The use of machine learning to discover novel materials and predict material properties has become a new research direction. On the basis of the current status of worldwide research on the combination of AI and crystal form and cocrystal, this mini-review analyzes and explores the application of AI in polymorphism prediction, crystal structure analysis, crystal property prediction, cocrystal former (CCF) screening, cocrystal composition prediction, and cocrystal formation prediction. This study provides insights into the future applications of AI in related fields.

Molecular, Solid-State and Surface Structures of the Conformational Polymorphic Forms of Ritonavir in Relation to their Physicochemical Properties

Purpose

Application of multi-scale modelling workflows to characterise polymorphism in ritonavir with regard to its stability, bioavailability and processing.

Methods

Molecular conformation, polarizability and stability are examined using quantum mechanics (QM). Intermolecular synthons, hydrogen bonding, crystal morphology and surface chemistry are modelled using empirical force fields.

Results

The form I conformation is more stable and polarized with more efficient intermolecular packing, lower void space and higher density, however its shielded hydroxyl is only a hydrogen bond donor. In contrast, the hydroxyl in the more open but less stable and polarized form II conformation is both a donor and acceptor resulting in stronger hydrogen bonding and a more stable crystal structure but one that is less dense. Both forms have strong 1D networks of hydrogen bonds and the differences in packing energies are partially offset in form II by its conformational deformation energy difference with respect to form I. The lattice energies converge at shorter distances for form I, consistent with its preferential crystallization at high supersaturation. Both forms exhibit a needle/lath-like crystal habit with slower growing hydrophobic side and faster growing hydrophilic capping habit faces with aspect ratios increasing from polar-protic, polar-aprotic and non-polar solvents, respectively. Surface energies are higher for form II than form I and increase with solvent polarity. The higher deformation, lattice and surface energies of form II are consistent with its lower solubility and hence bioavailability.

Conclusion

Inter-relationship between molecular, solid-state and surface structures of the polymorphic forms of ritonavir are quantified in relation to their physical-chemical properties.

Supplementary Information

The online version contains supplementary material available at 10.1007/s11095-021-03048-2.

Machine learning methods to predict the crystallization propensity of small organic molecules

Machine learning algorithms were explored for the prediction of the crystallization propensity based on molecular descriptors and fingerprints generated from 2D chemical structures and 3D chemical structures optimized with empirical methods.

A random forest model for predicting crystal packing of olanzapine solvates

A random forest (RF) classification model obtained from physicochemical properties of solvents and crystal structures of olanzapine has for the first time enabled the prediction of 3-D crystal packings of solvates. A novel solvate was obtained by targeted crystallization from the solvent identified by RF model.

A publicly available crystallisation data set and its application in machine learning

A publicly available crystallisation database for clusters of highly similar compounds is used to build machine learning models.

Can computed crystal energy landscapes help understand pharmaceutical solids?

Computational crystal structure prediction (CSP) methods can now be applied to the smaller pharmaceutical molecules currently in drug development. We review the recent uses of computed crystal energy landscapes for pharmaceuticals, concentrating on examples where they have been used in collaboration with industrial-style experimental solid form screening. There is a strong complementarity in aiding experiment to find and characterise practically important solid forms and understanding the nature of the solid form landscape.