Investigating Spatial Charge Descriptors for Prediction of Cocrystal Formation Using Machine Learning Algorithms

Abstract

Recently, drug modification via cocrystals has attracted great attention due to its high flexibility for the modulation of drug physicochemical properties. To reduce the cost of screening experiments, machine learning (ML) algorithms have proven to be one of the most effective ways to rapidly screen cocrystal formation. However, the choice of molecular descriptors has a significant impact on its prediction accuracy. In this work, two space-charge descriptors (COSMO-based σ-profile and three-dimensional (3D) spatial charge) are introduced into ML modeling, and two innovative COSMO-SVM and 3D-CNN ML algorithms are developed to screen the cocrystal formation. The two proposed ML models using space-charge descriptors demonstrate superior predictive accuracy for cocrystal formation compared to preproposed ML methods using extended connectivity fingerprints (ECFPs) and graph convolutional networks (GCNs). Considering computational efficiency and accuracy, 3D-CNN would be a very efficient ML model in the cocrystal formation screening task. In addition, the formation of two new cocrystals of norfloxacin was predicted by machine learning, and the cocrystals were obtained experimentally.

The Benefits and Challenges of Antibiotics–Non-Steroidal Anti-Inflammatory Drugs Non-Covalent Reaction

“…Another modeling application in multi-component research is the study of spatial charge descriptors, which can predict cocrystal formation using machine learning algorithms [144]. Two models were developed to predict the density of energetic and general organic cocrystals containing nitro groups, based on the artificial neural network (model I) and surface electrostatic potential correction method (model II), used to predict cocrystal density.…”

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

“…2,15−18 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. 19−22 Alternatively, various molecular descriptors have also been extracted and inputted in machine learning models such as multivariate adaptive regression splines, 23 multivariable logistic regression, 24 random forest (RF), 25 artificial neural network (ANN), 26 support vector machine, 22 and extreme gradient boosting. 12 In addition, various physics-inspired molecular representations, such as the Coulomb matrix (CM), bag-of-bonds (BoB), many-body tensor representation (MBTR), and smooth overlap of atomic positions (SOAP), have been proposed to capture different local and global molecular features.…”

Speeding Up the Cocrystallization Process: Machine Learning-Combined Methods for the Prediction of Multicomponent Systems

“…Moreover, the training set data should represent the samples to be predicted; otherwise, the ML model will have low predictive ability. Instead, as observed from the literature, 17,18,22 the features can be chosen from a plethora of descriptors and also the employed algorithms are several; however, the algorithm should be chosen coherently with the set of features and data. The purpose of this work is to develop a new predictive strategy and compare it with several methods already reported in the literature, emphasizing the advantages of the proposed approach and the points that make the prediction of new crystal forms challenging.…”

“…For these methods, it was estimated a variable accuracy in the range of 30–80% depending on the API . To overcome the poor accuracy of the property-based methods, a combination of different tools was also proposed, showing an improvement in the coformer selection of specific systems. ,− Recently, data-driven ML approaches have become increasingly popular due to the rapidity of calculation and promising predictive accuracy. − Several algorithms were evaluated, such as support vector machine (SVM), random forest (RF), neural networks (NN), and partial least squares-discriminant analysis (PLS-DA), and also, a wide variety of molecular representations were considered, including molecular descriptors, fingerprint vectors, and molecular graphs . To mention a few studies, Fornari et al proposed using QSAR descriptors and the PLS-DA model to discriminate between the formation of cocrystals and physical mixtures .…”

“…Moreover, the training set data should represent the samples to be predicted; otherwise, the ML model will have low predictive ability. Instead, as observed from the literature, ,, the features can be chosen from a plethora of descriptors and also the employed algorithms are several; however, the algorithm should be chosen coherently with the set of features and data.…”