On the Development of Descriptor-Based Machine Learning Models for Thermodynamic Properties: Part 1—From Data Collection to Model Construction: Understanding of the Methods and Their Effects

Abstract

In the present work, a multi-angle approach is adopted to develop two ML-QSPR models for the prediction of the enthalpy of formation and the entropy of molecules, in their ideal gas state. The molecules were represented by high-dimensional vectors of structural and physico-chemical characteristics (i.e., descriptors). In this sense, an overview is provided of the possible methods that can be employed at each step of the ML-QSPR procedure (i.e., data preprocessing, dimensionality reduction and model construction) and an attempt is made to increase the understanding of the effects related to a given choice or method on the model performance, interpretability and applicability domain. At the same time, the well-known OECD principles for the validation of (Q)SAR models are also considered and addressed. The employed data set is a good representation of two common problems in ML-QSPR modeling, namely the high-dimensional descriptor-based representation and the high chemical diversity of the molecules. This diversity effectively impacts the subsequent applicability of the developed models to a new molecule. The data set complexity is addressed through customized data preprocessing techniques and genetic algorithms. The former improves the data quality while limiting the loss of information, while the latter allows for the automatic identification of the most important descriptors, in accordance with a physical interpretation. The best performances are obtained with Lasso linear models (MAE test = 25.2 kJ/mol for the enthalpy and 17.9 J/mol/K for the entropy). Finally, the overall developed procedure is also tested on various enthalpy and entropy related data sets from the literature to check its applicability to other problems and competing performances are obtained, highlighting that different methods and molecular representations can lead to good performances.

On the Development of Descriptor-Based Machine Learning Models for Thermodynamic Properties: Part 2—Applicability Domain and Outliers

“…This observation, in conjunction with the distribution of enthalpy values in the dataset (cf. the first article of the series [15]), corroborates the hypothesis of poor representation of these molecules in the dataset (e.g., very large molecules). In this case, the elimination of these points is questionable.…”

“…Then, the resulting preprocessed data without outliers were split into training and test sets, the ratio between them being fixed at 80:20, and scaled via a standard scaling method. This ratio for data splitting and this scaling method were indeed identified as well performing in the first article of the series [15]. To better integrate the effect of data splitting on the performance of the models, five different training/test splits were considered.…”

“…In the previous article, two ML-QSPR models were built for the ideal gas phase standard enthalpy of formation and entropy, based on data from the DIPPR database [15]. These properties will be denoted as enthalpy (H) and entropy (S) in the rest of this article, for simplicity.…”

“…Furthermore, a dimensionality reduction step is implemented to improve interpretability of the developed ML models. To achieve this, the same GA procedure as the one used in the first article is used to reduce the numbers of descriptors to a fixed number of 100 [15]. Finally, the Lasso models are trained and validated on the new dataset of reduced dimension.…”

“…This article is the second of a series of articles aiming to construct QSPR models on the basis of ML techniques, ML-QSPR, while examining the different choices that are offered to the developer along the way and analysing the effects of each one. In the first article of the series, two ML-QSPR models were developed to predict two thermodynamic properties, namely the enthalpy of formation (H) and the absolute entropy (S) for the ideal gas state of molecules at 298.15 K and 1 bar [15]. The molecular descriptors were employed as features of the models to describe the structural and atomic characteristics of the considered molecules of the DIPPR (Design Institute for Physical Properties) database.…”