The effect of descriptor choice in machine learning models for ionic liquid melting point prediction

Abstract

The characterization of an ionic liquid’s properties based on structural information is a longstanding goal of computational chemistry, which has received much focus from ab initio and molecular dynamics calculations. This work examines kernel ridge regression models built from an experimental dataset of 2212 ionic liquid melting points consisting of diverse ion types. Structural descriptors, which have been shown to predict quantum mechanical properties of small neutral molecules within chemical accuracy, benefit from the addition of first-principles data related to the target property (molecular orbital energy, charge density profile, and interaction energy based on the geometry of a single ion pair) when predicting the melting point of ionic liquids. Out of the two chosen structural descriptors, ECFP4 circular fingerprints and the Coulomb matrix, the addition of molecular orbital energies and all quantum mechanical data to each descriptor, respectively, increases the accuracy of surrogate models for melting point prediction compared to using the structural descriptors alone. The best model, based on ECFP4 and molecular orbital energies, predicts ionic liquid melting points with an average mean absolute error of 29 K and, unlike group contribution methods, which have achieved similar results, is applicable to any type of ionic liquid.

A review on machine learning algorithms for the ionic liquid chemical space

“…al. 92 (2020) investigated the effect of descriptor choice on melting point prediction. They used Venkatraman's dataset of 2200 ILs and quantum mechanical descriptors.…”

“…very accurately state in their work that many semi-empirical predictions could likely be rened by using a higher level of theory during initial parameter selection, instead of using the arbitrarily-engineered features that are popular in many models. 92 In practise this would mean choosing IL descriptors that are based on distinctive properties (such as HOMO-LUMO gap, or s-proles) instead of an articial representation that has no meaning in physical space (such as SMILES descriptors). On the other hand, explaining the parameters of non-linear correlations, such as those easily detected by NNs is far more difficult.…”

Optimal Aqueous Biphasic Systems Design for the Recovery of Ionic Liquids

“…Unfortunately, thermodynamic models that can provide such predictions are still not available due to the high complexity of these biphasic systems. On the other hand, machine learning (ML) algorithms such as ANN and SVM have been employed to build complex nonlinear GC or QSPR models for different properties such as gas solubility, − surface tension, viscosity, toxicity, , melting point, , and the acid dissociation constants of organic compounds . Besides these, some of ML-based models have been integrated into the computer-aided design method for addressing some optimal design problems such as CO 2 capture , and cosmetic formulation .…”

Surface tension of binary mixtures containing environmentally friendly ionic liquids: Insights from artificial intelligence

“…Among these interesting and innovative methods, the attention of some researchers has been drawn to AI models in order to predict the different properties of pure (Deng et al, 2020;Lazzús, 2017;Low et al, 2020;Mulero et al, 2017;Wang et al, 2021), binary (Lashkarbolooki, 2017;Lashkarbolooki et al 2012Lashkarbolooki et al , 2013, and ternary (Hezave et al 2013) systems containing IL.…”