QSPR modeling of selectivity at infinite dilution of ionic liquids

The intelligent choice of extractants and entrainers can improve current mixture separation techniques allowing better efficiency and sustainability of chemical processes that are both used in industry and laboratory practice. The most promising approach is a straightforward comparison of selectivity at infinite dilution between potential candidates. However, selectivity at infinite dilution values are rarely available for most compounds so a theoretical estimation is highly desired. In this study, we suggest a Quantitative Structure–Property Relationship (QSPR) approach to the modelling of the selectivity at infinite dilution of ionic liquids. Additionally, auxiliary models were developed to overcome the potential bias from big activity coefficient at infinite dilution from the solute. Data from SelinfDB database was used as training and internal validation sets in QSPR model development. External validation was done with the data from literature. The selection of the best models was done using decision functions that aim to diminish bias in prediction of the data points associated with the underrepresented ionic liquids or extreme temperatures. The best models were used for the virtual screening for potential azeotrope breakers of aniline + n-dodecane mixture. The subject of screening was a combinatorial library of ionic liquids, created based on the previously unused combinations of cations and anions from SelinfDB and the test set extractants. Both selectivity at infinite dilution and auxiliary models show good performance in the validation. Our models’ predictions were compared to the ones of the COSMO-RS, where applicable, displaying smaller prediction error. The best ionic liquid to extract aniline from n-dodecane was suggested.

Catalytic Performance of Cycloalkyl-Fused Aryliminopyridyl Nickel Complexes toward Ethylene Polymerization by QSPR Modeling

Quantitative structure–property relationship (QSPR) modeling is performed to investigate the role of cycloalkyl-fused rings on the catalytic performance of 46 aryliminopyridyl nickel precatalysts. The catalytic activities for nickel complexes in ethylene polymerization are well-predicted by the obtained 2D-QSPR model, exploring the main contribution from the charge distribution of negatively charged atoms. Comparatively, 3D-QSPR models show better predictive and validation capabilities than that of 2D-QSPR for both catalytic activity (Act.) and the molecular weight of the product (Mw). Three-dimensional contour maps illustrate the predominant effect of a steric field on both catalytic properties; smaller sizes of cycloalkyl-fused rings are favorable to Act.y, whereas they are unfavorable to Mw. This study may provide assistance in the design of a new nickel complex with high catalytic performance.

QSPR MODELS FOR ZETA-POTENTIAL OF NANO-OXIDES PREDICTION

Nano-QSPR modeling often requires considering variety of factors, if neglected, may lead to erroneous result of the study. Frequently, the data turned out to be inaccurate, incomplete, or fragmentary. Obviously, the quality of experimental data directly depends on many factors: laboratory equipment, organization of internal regulations, skills of researchers, and so on. As a result of violations of algorithms and protocols of initial data streams processing – there are errors and distortions of data, that is why performing a solid multistep data-curation process is crucial for such procedures. Data curation procedure was performed and approximately 60% was rejected (due to various errors, incomplete or absent records for physicochemical parameters or conditions of performed experiment), followed up by using zeta-potential value dataset for 37 various sizes nanoparticles of 14 metal oxides for calculation of 1D SiRMS descriptors as well as «liquid drop» model cross-descriptors. An efficient consensus model was built (R2 = 0.88, R2test = 0.81). Predictive power (R2 = 0.84) of the model was tested using an external set of 5 nano-oxides and the possibility of satisfactory zeta-potential prediction was shown. Prediction of zeta-potential value within domain applicability of obtained QSPR model confirmed using a Williams plot. The interpretation of the final model was carried out and it was found that the contribution of descriptors was distributed between individual descriptors and cross-descriptors by 46% and 54% respectively. The contribution 1D SiRMS descriptors was 59%, the second group – 41% (liquid drop model descriptors – 29%, descriptors characterizing the metal atom – 12%). It was found that the most influential parameters are the characteristics that reflect the nature of the oxides. The parameters of electrostatic interactions have the highest contribution.

A systematic DNN-based QSPR modeling methodology for rapid and reliable prediction on flashpoints of chemicals

Quantitative structure-property relationship (QSPR) studies based on deep neural networks (DNN) are receiving increasing attention due to their excellent performances. A systematic methodology coupling multiple machine learning technologies is proposed to solve vital problems including applicability domain and prediction uncertainty in DNN-based QSPRs. Key features are rapidly extracted from plentiful but chaotic descriptors by principal component analysis (PCA) and kernel PCA. Then, a detailed applicability domain (AD) is defined by K-means algorithm to avoid unreliable predictions and discover its potential impact on uncertainty. Moreover, prediction uncertainty is analyzed with dropout-embedded DNN by thousands of independent tests to assess the reliability of predictions. The prediction of flashpoint temperature is employed as a case study demonstrating that the model accuracy is remarkably improved comparing with the referenced model. More importantly, the proposed methodology breaks through difficulties in analyzing the uncertainty of DNN-based QSPRs and presents an AD correlated with the uncertainty.