Predicting the Mixing Behavior of Aqueous Solutions Using a Machine Learning Framework

Abstract

The most direct approach to determining if two aqueous solutions will phase-separate upon mixing is to exhaustively screen them in a pair-wise fashion. This is a timeconsuming process that involves preparation of numerous stock solutions, precise transfer of highly concentrated and often viscous solutions, exhaustive agitation to ensure thorough mixing, and time-sensitive monitoring to observe the presence of emulsion characteristics indicative of phase separation. Here, we examined the pair-wise mixing behavior of 68 water-soluble compounds by observing the formation of microscopic phase boundaries and droplets of 2278 unique 2-component solutions. A series of machine learning classifiers (artificial neural network, random forest, k-nearest neighbors, and support vector classifier) were then trained on physicochemical property data associated with the 68 compounds and used to predict their miscibility upon mixing. Miscibility predictions were then compared to the experimental observations. The random forest classifier was the most successful classifier of those tested, displaying an average receiver operator characteristic area under the curve of 0.74. The random forest classifier was validated by removing either one or two compounds from the input data, training the classifier on the remaining data and then predicting the miscibility of solutions involving the removed compound(s) using the classifier. The accuracy, specificity, and sensitivity of the random forest classifier were 0.74, 0.80, and 0.51, respectively, when one of the two compounds to be examined was not represented in the training data. When asked to predict the miscibility of two compounds, neither of which were represented in the training data, the accuracy, specificity, and sensitivity values for the random forest classifier were 0.70, 0.82 and 0.29, respectively. Thus, there is potential for this machine learning approach to improve the design of screening experiments to accelerate the discovery of aqueous two-phase systems for numerous scientific and industrial applications.

Data-Driven Imputation of Miscibility of Aqueous Solutions via Graph-Regularized Logistic Matrix Factorization

“…To test our approach, we used the experimental data from Peacock et al This data set provides (i) the complete miscibility matrix M complete containing all pairwise mixing outcomes of n = 68 aqueous solutions of distinct compounds at compound-specific concentrations (1559 miscible and 719 immiscible pairs) near room temperature and (ii) for the solution graph G , the category to which each compound belongs: 46 polymers, 11 surfactants, 8 proteins, and 3 salts. See Figure S2 for the complete miscibility matrix M complete for all n ( n – 1)/2 = 2278 pairs of solutions.…”

“…As a vector representation of a given solution of a compound, we use (i) the physicochemical features compiled from PubChem by Peacock et al: monomer and polymer molecular weight, the log of the predicted octanol–water partition coefficient, hydrogen bond donor and acceptor counts, and complexity, (ii) the concentration of the compound in the solution, and (iii) a one-hot encoding of the category (polymer, protein, surfactant, salt) of its solute. Note, since the compounds could not be fully annotated with group (i) features, we used the imputed features based on an 8-nearest-neighbors algorithm from Peacock et al This gives a length-11 feature vector for each solution. To represent a pair of solutions, we concatenate the vector representations of the two compounds, giving a 22-dimensional input to the RF.…”

“…All code to reproduce our plots in is available at . The raw miscibility data is also on our Github repo in comma-separated-value () format, which we reformatted from the Supporting Information of Peacock et al The required calculations are modest for this scale of data; on a 2020 Macbook Pro M1 laptop, training a single GR-LMF model requires less than 1 s of runtime, and our entire hyperparameter sweep requires less than 1 min.…”

Can Machine Learning Predict the Phase Behavior of Surfactants?

“…ML has also been applied to phase diagram prediction. − For example, a ML force field has been used to compute the phase diagram for uranium for temperatures and pressures up to 1600 K and 800 GPa, respectively, with relatively good accuracy at lower temperatures but diverging from density functional theory predictions at higher temperatures and pressures . That work differs from that presented here, in that we are looking to compute the phase diagram directly from a data set of experimentally determined phase diagrams, as opposed to predicting the phase indirectly using a ML force field.…”

Machine learning‐assisted multiscale modeling of an autothermal fixed‐bed reactor for methanol to propylene process

“…Once trained with known data, this black‐box type model is intended to estimate the output accurately when new input variables are introduced. With the current resurgence of interest in data‐driven models, neural network‐based machine learning techniques have been harnessed in various purpose, property estimation, 29 flow regime recognition, 30 reactor sensitivity analysis, 31 and so on. Note that massive data generated from physics‐based model or experiments is required to train the neural network, but the data at reactor scale is not available for the present work.…”