A paper from the Laboratory of Data and Chemical Engineering (headed by the Associate Professor KANEKO Hiromasa) graced the cover of Volume 63, Issue 1 (2024) of the international academic journal Industrial & Engineering Chemistry Research.
In the development of pharmaceuticals and agrochemicals, machine learning is attracting attention for accelerating the drug discovery process, and machine learning models are being used to predict the pharmacological activity of molecules. To build a model with high prediction accuracy, proper consideration of the molecule's three-dimensional structure is essential, but conventional methods require a common skeletal structure for the group of compounds used to construct the model. In this research, a molecular descriptor that takes into account the three-dimensional characteristics of chemical structures, which can be employed even when a group of the compounds lacks a common backbone structure, was developed. In the proposed method, descriptors are calculated by combining information on a positive charge, negative charge, surface, and hydrophobicity obtained using Flare, a molecular design software, with k-means++, a clustering method, and a Gaussian mixture model. The calculated descriptors are highly interpretable, offering insights into critical structures related to electrical properties, steric properties, and hydrophobicity based on the significance of the clusters. The proposed descriptors' validity and interpretability were confirmed through verification using datasets for herbicides, angiotensin-converting enzyme inhibitors, ecotoxicity, and water solubility.
Supplementary Cover Art:
Industrial & Engineering Chemistry Research
Article:
Development of New Molecular Descriptors Based on Flare Software Considering Three-Dimensional Chemical Structures
≪
Japanese version≫
