Discussion on Regression Methods Based on Ensemble Learning and Applicability Domains of Linear Submodels

Evaluation and Optimization Methods for Applicability Domain Methods and Their Hyperparameters, Considering the Prediction Performance of Machine Learning Models

In molecular, material, and process design and control, the applicability domain (AD) of a mathematical model y = f(x) between properties, activities, and features x is constructed. As there are multiple AD methods, each with its own set of hyperparameters, it is necessary to select an appropriate AD method and hyperparameters for each data set and mathematical model. However, there is no method for optimizing the AD model. This study proposes a method for evaluating and optimizing the AD model for each data set and a mathematical model. Using the predictions of double cross-validation with all samples, the relationship between coverage and root-mean-squared error (RMSE) was calculated for all combinations of AD methods and their hyperparameters, and the area under the coverage and RMSE curve (AUCR) was calculated. The AD model with the lowest AUCR value was selected as the optimal fit for the mathematical model. The proposed method was validated using eight data sets, including molecules, materials, and spectra, demonstrating that the proposed method could generate optimal AD models for all data sets. The Python code for the proposed method is available at https://github.com/hkaneko1985/dcekit.

Combining Ensemble Learning with a Fragment-Based Topological Approach To Generate New Molecular Diversity in Drug Discovery: In Silico Design of Hsp90 Inhibitors

Machine learning methods have revolutionized modern science, providing fast and accurate solutions to multiple problems. However, they are commonly treated as "black boxes". Therefore, in important scientific fields such as medicinal chemistry and drug discovery, machine learning methods are restricted almost exclusively to the task of performing predictions of large and heterogeneous data sets of chemicals. The lack of interpretability prevents the full exploitation of the machine learning models as generators of new chemical knowledge. This work focuses on the development of an ensemble learning model for the prediction and design of potent dual heat shock protein 90 (Hsp90) inhibitors. The model displays accuracy higher than 80% in both training and test sets. To use the ensemble model as a generator of new chemical knowledge, three steps were followed. First, a physicochemical and/or structural interpretation was provided for each molecular descriptor present in the ensemble learning model. Second, the term "pseudolinear equation" was introduced within the context of machine learning to calculate the relative quantitative contributions of different molecular fragments to the inhibitory activity against the two Hsp90 isoforms studied here. Finally, by assembling the fragments with positive contributions, new molecules were designed, being predicted as potent Hsp90 inhibitors. According to Lipinski's rule of five, the designed molecules were found to exhibit potentially good oral bioavailability, a primordial property that chemicals must have to pass early stages in drug discovery. The present approach based on the combination of ensemble learning and fragment-based topological design holds great promise in drug discovery, and it can be adapted and applied to many different scientific disciplines.