Cocrystal Prediction of Bexarotene by Graph Convolution Network and Bioavailability Improvement

Abstract

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.

Multi-type Features Graph Neural Networks for Cocrystals Prediction using API-Coformers Interactions

“…In the experiment finding the suitable architecture for compelling results, on trials, we tried various number convolution layers and ultimately found out that two convolution layers were promising in terms of accuracy, sensitivity, specificity, and MCC compared to the other architectures as shown in Table 3. This architecture seems to have performed better in various fields specifically cocrystal engineering, drug and protein interactions predictions 12,48,51,53 . It was observed that in Table 3 our GNN models with few layers (i.e., two layers) gave the best performance and did not improve the predictive performance and tended to perform increasingly worse on classifying graph nodes as the model gets deeper (i.e., piled up many layers and add non-linearity).…”

“…It is paramount to develop complementary tools that can shorten the list selection of coformers by predicting which coformers are most likely to form cocrystals successfully. The experimental determination and traditional screening methods for cocrystals such as solution crystallization, liquid-assisted and dry grinding, and anti-solvent addition are tremendously time-consuming, economically expensive, and labor-intensive 11,12 .…”

Machine Learning-Guided Prediction of Cocrystals Using Point Cloud-Based Molecular Representation

“…Graph convolutional network (GCN) is by far the most common machine learning algorithm used for cocrystal prediction, in which molecules are represented as a set of matrices to capture atom connectivity and features. − Alternatively, various molecular descriptors have also been extracted and inputted in machine learning models such as multivariate adaptive regression splines, multivariable logistic regression, random forest (RF), artificial neural network (ANN), support vector machine, and extreme gradient boosting …”

Leveraging Deep Chemical Language Processing for Co-crystal Prediction

“…The early approaches adopt either fully connected neural networks or tree-based models on molecular descriptors to predict the cocrystallization of molecule pairs [21,22,24]. More recent works introduced deep learning to the task and utilized convolutions over molecular graphs [10,23] to improve the predictive performance. Common to those models, they are trained on unbalanced datasets with folds of more API-coformer co-crystals than negative pairs, due to limited data availability, and struggle to generalize unseen data [21].…”