TY  - JOUR
AB  - The working temperature range of Ionic Liquids (IL) is determined by their liquid state range, where the IL's melting point/glass transition and decomposition temperatures define the lower and upper limits of the range, respectively. Computational prediction of the structure of new ILs with required properties, e.g. which can exist in a liquid state at room temperature and are stable up to high temperatures, is a much less time-consuming and less expensive approach than stepwise synthesis and experimental examination of all presupposed ILs. Therefore, in the present work the quantitative structure–property relationship (QSPR) models were developed to predict the glass transition temperature (Tg), melting point (Tm), and decomposition temperatures (Td) of ILs. We showed that a use of component validation protocol provided a better agreement of statistical parameters for the training and test sets. The performance of various modeling algorithms and descriptor sets was discussed and compared and advantages of descriptor-less as well as multi-task modeling were shown. An explanation of the models using statistical analysis of functional groups and Molecular Matched Pairs were provided for the mixtures for the first time. The experimental data and models, which are the first publicly available models for prediction of transition and decomposition temperatures of ILs, are publicly available online at http://ochem.eu/article/140250.
AU  - Makarov, D.M.*
AU  - Fadeeva, Y.A.*
AU  - Shmukler, L.E.*
AU  - Tetko, I.V.
C1  - 66109
C2  - 53089
TI  - Machine learning models for phase transition and decomposition temperature of ionic liquids.
JO  - J. Mol. Liq.
VL  - 366
PY  - 2022
SN  - 0167-7322
ER  - 

TY  - JOUR
AB  - The melting point (MP) of an ionic liquid (IL) is one of the key physical properties as it determines the lower limit of the IL working temperature range. In this work, we analysed the recently published studies to predict MP of ILs. While we were able to reproduce the statistical parameters reported by the authors, we found that the performance of the models with new test set data was much lower than the reported statistical values. The discrepancy was due to the validation protocol (random split of the initial set into training/test subsets) that did not allow correct estimation of how contributions of individual ions affect the model performance. Using a more rigorous validation protocol we reached good agreement between the training and test set statistical parameters. We strongly suggest using this protocol for proper validation of models for other properties of ILs to avoid reporting overoptimistic statistical parameters. We also showed that the Transformer Convolutional Neural Network, which was based on the representation of molecules as text (SMILES), proposed a model with significantly higher prediction accuracy as compared to those developed using descriptors that were used in the previous studies. The RMSE of this model is 44 °C and the model is applicable to any type of ILs. The data and developed models are publicly available online at http://ochem.eu/article/135195.
AU  - Makarov, D.M.*
AU  - Fadeeva, Y.A.*
AU  - Shmukler, L.E.*
AU  - Tetko, I.V.
C1  - 63256
C2  - 51410
CY  - Radarweg 29, 1043 Nx Amsterdam, Netherlands
TI  - Beware of proper validation of models for ionic Liquids!
JO  - J. Mol. Liq.
VL  - 344
PB  - Elsevier
PY  - 2021
SN  - 0167-7322
ER  -