Solubility Prediction from Molecular Properties and Analytical Data Using an In-phase Deep Neural Network (Ip-DNN)

Area of Science:

• Materials Science
• Computational Chemistry
• Data Science

Background:

• Materials informatics accelerates research by predicting material properties.
• Solubility parameters like Hansen and Hildebrand solubility parameters (HSPs) and Log P are crucial for understanding substance behavior.
• Accurate solubility prediction is vital across various scientific and industrial applications.

Purpose of the Study:

• To develop a novel, accurate solubility prediction tool.
• To leverage machine learning for predicting solubility using only analytical input data.
• To make the developed tool accessible to the scientific community.

Main Methods:

• Utilized a unique in-phase deep neural network (ip-DNN) machine learning approach.
• Employed analytical data (NMR, refractive index, density) as direct input.
• Incorporated intermediate regression models to enhance prediction accuracy within the ip-DNN framework.

Main Results:

• Successfully established a functional solubility prediction tool.
• Demonstrated the efficacy of the ip-DNN method for predicting solubility parameters.
• Achieved improved prediction accuracy through intermediate regression models.

Conclusions:

• The developed ip-DNN method provides a powerful approach for predicting material solubility.
• The tool, accessible via a dedicated website, facilitates materials science research.
• This work advances the application of machine learning in materials informatics.