Hybridizing physical and data-driven prediction methods for physicochemical properties

Jirasek, Fabian; Bamler, Robert; Mandt, Stephan

doi:10.1039/d0cc05258b

Cited by 20 publications

(23 citation statements)

References 24 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The hybrid GNN models further improve this performance in all cases except for the UNIFAC models. Similar behavior was encountered by the comparison between bagging and boosting models for UNIFAC-Dortmund using the matrix completion method- 40 . This demonstrates that, in general, by combining both mechanistic and data-driven models more reliable predictions are achieved.…”

Section: Hybrid Gnn Modelsupporting

confidence: 61%

“…In this approach many quantum-chemical and/or topological descriptors are calculated and discarded for not being relevant enough for an accurate prediction. A notable exception to this was the application of the matrix completion methodology (MCM) that allows for the predictions to be based only on an incomplete matrix of solvent-solute activity coefficients 21,40,41 . This eliminates the necessity of computing many expensive descriptors.…”

Section: Introductionmentioning

confidence: 99%

“…A notable exception to this was the application of the matrix completion methodology (MCM) that allows for the predictions to be based only on an incomplete matrix of solvent-solute activity coefficients. 21,40,41 This eliminates the necessity of computing many expensive descriptors. However, the application domain of the MCM method is limited to the systems that dene the matrix.…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Graph neural networks for the prediction of infinite dilution activity coefficients

et al. 2022

View full text Add to dashboard Cite

show abstract

Section: Hybrid Gnn Modelsupporting

confidence: 61%

Section: Introductionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Graph neural networks for the prediction of infinite dilution activity coefficients

et al. 2022

View full text Add to dashboard Cite

show abstract

“…It was also shown that it is possible to combine the Bayesian MCM with UNIFAC. 9 In this work, we use an MCM based on iterative principal component analysis (PCA), which is an alternative to the Bayesian MCM. The focus of the present work is on temperature-dependent limiting activity coefficients.…”

Section: Introductionmentioning

confidence: 99%

“…Jirasek et al have used a Bayesian matrix completion method (MCM) for predicting limiting activity coefficients in binary mixtures at 298.15 K. In contrast to the UNIFAC and COSMO-RS prediction methods, the predictions obtained with this MCM are made without using any physical knowledge on the components of the mixtures. It was also shown that it is possible to combine the Bayesian MCM with UNIFAC …”

Section: Introductionmentioning

confidence: 99%

Predicting Activity Coefficients at Infinite Dilution for Varying Temperatures by Matrix Completion

Damay

Jirasek

Kloft

et al. 2021

Ind. Eng. Chem. Res.

Self Cite

View full text Add to dashboard Cite

Activity coefficients describe the nonideality of liquid mixtures and are essential for calculating equilibria. The activity coefficients at infinite dilution in binary mixtures are particularly important as the activity coefficients at finite concentrations can be predicted based on their knowledge not only in binary mixtures but also in multicomponent mixtures. The available experimental data on these activity coefficients at infinite dilution in binary mixtures is readily accessible in databases and can be organized in a matrix with the rows representing the solutes and the columns representing the solvents or vice versa. As experimental data is lacking for many binary mixtures, this matrix is only sparsely populated. Filling its gaps using predictive methods is essential. Recently, matrix completion methods (MCMs) have been applied successfully for this purpose. However, only isothermal data sets have been considered. In the present work, we apply an MCM to predict activity coefficients at infinite dilution at varying temperatures. Furthermore, we show how one can incorporate physical knowledge on the nature of the temperature dependency of the activity coefficients at infinite dilution. The predictions obtained with this new approach outperform those obtained with the best currently available physical prediction method for activity coefficients at infinite dilution, the modified UNIFAC (Dortmund) method.

show abstract

Prediction of Henry's law constants by matrix completion

2022

Self Cite

View full text Add to dashboard Cite

Methods for predicting Henry's law constants H ij are important as experimental data are scarce. We introduce a new machine learning approach for such predictions: matrix completion methods (MCMs) and demonstrate its applicability using a data base that contains experimental H ij values for 101 solutes i and 247 solvents j at 298 K. Data on H ij are only available for 2661 systems i + j. These H ij are stored in a 101 Â 247 matrix; the task of the MCM is to predict the missing entries. First, an entirely data-driven MCM is presented. Its predictive performance, evaluated using leave-one-out analysis, is similar to that of the Predictive Soave-Redlich-Kwong equation-of-state (PSRK-EoS), which, however, cannot be applied to all studied systems. Furthermore, a hybrid of MCM and PSRK-EoS is developed in a Bayesian framework, which yields an unprecedented performance for the prediction of H ij of the studied data set.

show abstract

Hybridizing physical and data-driven prediction methods for physicochemical properties

Abstract: We present a generic way to hybridize physical and data-driven methods for predicting physicochemical properties. The approach `distills' the physical method's predictions into a prior model and combines it with...

Cited by 20 publications

References 24 publications

Graph neural networks for the prediction of infinite dilution activity coefficients

Graph neural networks for the prediction of infinite dilution activity coefficients

Predicting Activity Coefficients at Infinite Dilution for Varying Temperatures by Matrix Completion

Prediction of Henry's law constants by matrix completion

Contact Info

Product

Resources

About