No abstract

Mechanochemistry is a green preparation method that uses mechanical forces to prompt chemical reactions. This technique has shown its potential as an efficient alternative for several solvent-based processes (e.g., synthesis of co-crystals, metal complexes, or polymers); however, predicting its reactivity remains a challenge. In this study, a machine learning model was developed to gain insights into this process and predict the formation of co-amorphous mixtures. Co-amorphous mixtures are produced when the molecular arrangement of a crystalline active pharmaceutical ingredient is disrupted and maintained at "random" by the synergistic presence of a secondary structure. Co-amorphous mixtures can be designed as multicomponent drugs and often display an enhanced solubility and bioavailability. In this work, we generated a database of 418 inhouse amorphization experiments, novel to current literature, and informed data analysis (i.e., gradient boosting and neural networks) for predictive purposes and to extrapolate fundamental insights. By using 2066 chemical descriptors to train a gradient boost model, a predictive accuracy of >73% was achieved. This model was further used to predict and synthesize six novel co-amorphous mixtures. We expect that this novel database and the predictive model will aid at designing novel pharmaceuticals and advancing sustainable solvent-free processes.

No abstract

No abstract

The amorphization of pharmaceutical crystals is an effective strategy to enhance the bioavailability of poorly soluble active pharmaceutical ingredients (APIs). However, this process can be challenging as these supramolecular structures...