Solvate Prediction for Pharmaceutical Organic Molecules with Machine Learning

Abstract

Methods to predict crystallization behavior for active pharmaceutical ingredients (APIs) can serve as an important guide in small molecule pharmaceutical development. Here, we describe solvate formation propensity prediction for pharmaceutical molecules via a machine learning approach. Random forests (RF) and support vector machine (SVM) algorithms were trained and tested with data sets extracted from Cambridge Structural Database (CSD). The machine learning models, requiring only 2D structures as input, were able to predict solvate formation propensity for organic molecules with up to 86% success rate. Performance of the models was demonstrated with a collection of 20 pharmaceutical molecules.

One class classification as a practical approach for accelerating π–π co-crystal discovery

“…This means that, from a machine learning perspective, only one class (i.e., the positive outcome) is well dened by the data. Recent research from a range of groups has attempted to tackle this unbalanced data problem for prediction problems like cocrystallization, 11,40 solvate formation 80,81 and crystallisability. 82,83 In general, these groups have attempted to get around the problem by using either sparse or somewhat unreliable negative data from alternative sources to produce a trained model.…”

Crystalline Multicomponent Solids: An Alternative for Addressing the Hygroscopicity Issue in Pharmaceutical Materials

“…69 Xin and co-workers made use of random forests and support vector machine learning algorithms to develop models for predicting the solvate formation propensity of organic molecules. 70 Crystal engineering plays a pivotal role in improving many physicochemical properties of pharmaceutical solids. Various novel single or multicomponent solids can be prepared using crystal engineering approaches that include preparation of polymorphs, solvates/hydrates, pharmaceutical salts, cocrystals, as well as eutectics to alter various physicochemical properties.…”

New insights and innovation from a million crystal structures in the Cambridge Structural Database

“…Many research projects are benefitting from API access: For example, users have been able to more easily use machine learning in tandem with the API for solvate prediction, 38 to help implement fragment pocket analysis using structural informatics, 39 and to aid with crystal structure prediction, 29 for understanding of the impact of compression of cocrystals 40 (of interest in the formation of tablets) and for parametrization of structural refinement programs 41 …”