(565e) Prediction of the Formation of Co-Amorphous Systems By Means of a Machine Learning Approach | AIChE

(565e) Prediction of the Formation of Co-Amorphous Systems By Means of a Machine Learning Approach

Authors 

Fink, E. - Presenter, Research Center Pharmaceutical Engineering Gmbh
Brunsteiner, M., Research Center Pharmaceutical Engineering GmbH
Paudel, A., Institute of Process and Particle Engineering, Graz University of Technology
Zellnitz-Neugebauer, S., Research Center Pharmaceutical Engineering GmbH
Amorphous forms and formulations of drugs have gained interest due to several favorable properties, but compared to their crystalline counterparts one disadvantage is their thermodynamic instability. Co-amorphous systems (COAMS) have been described as a promising solution to overcome this drawback and stabilize the amorphous form (Shi et al., 2019). The co-former selection is a crucial factor, however, a lack of systematic, predictive and computational methods for co-former selection has been identified (Liu et al., 2021). So far, co-former selection has mainly been done case-by-case, based on synergistic effects and/or combination therapy, previous studies, physicochemical properties or structural analysis.

This work presents a machine learning (ML) approach to predict the successful formation of new COAMS. A model based on gradient boosting methods was developed on a training dataset comprising 254 systems described in literature. A careful selection of molecular descriptors was used to obtain a numerical description of the systems. In order to increase the robustness and generalizability of our model, an ensemble learning approach was chosen in combination with cross-validation. On the training data, a correct prediction was observed in 80% of the cases and on the validation data (21 additional systems) on 76%.

Further, the model was used to predict the formation of COAMS in new, unreported systems. These consisted of two APIs as used in the field of inhalation therapy. Using our model and analysis of the training data, promising combinations of APIs, possibly exhibiting co-amorphous properties, were identified based on their likelihood of forming a COAMS. This likelihood was estimated based on the distance of the new system from the original training data and the predicted class (COAMS or not). Several of the suggested COAMS were further tested to experimentally validate the model. The developed model will expedite the experimental screening phase for new co-amorphous systems and help saving time and cost.


References:

Liu, J., Grohganz, H., Löbmann, K., Rades, T., Hempel, N.J., 2021. Co-amorphous drug formulations in numbers: Recent advances in co-amorphous drug formulations with focus on co-formability, molar ratio, preparation methods, physical stability, in vitro and in vivo performance, and new formulation strategies. Pharmaceutics 13. https://doi.org/10.3390/pharmaceutics13030389

Shi, Q., Moinuddin, S.M., Cai, T., 2019. Advances in coamorphous drug delivery systems. Acta Pharm. Sin. B 9, 19–35. https://doi.org/10.1016/j.apsb.2018.08.002