The Hildebrand solubility is a physiochemical parameter which is widely accepted as a viable estimate of interactive forces responsible for compatibility between compounds. The Hansen solubility parameters (HSP) are an extension of this theory where the Hildebrand solubility is divided into three components: dispersion, polarity, and hydrogen bonding. HSP determines solvent-polymer interactions based on the principle of ‘like dissolves like’. The Euclidean distance between the HSP of two compounds is a measure of ‘likeness’. This is a highly useful method to describe solvent-polymer, nanoparticle-dispersion, and liquid-liquid interactions. Both experimental and computational methods to calculate HSPs are time-consuming and expensive. Our group’s previous attempts to predict HSP using simple machine learning algorithms yielded promising results. A comprehensive feature analysis yielded the best feature set combinations for each HSP (diffusion, polarity, and hydrogen bonding). Using these insights we train advanced deep neural network architectures to predict Hildebrand solubilities of 1167 well-known organic solvents. Here, we present two deep learning models. The first model is a multi-task approach to predict all three HSPs using one deep neural network. The second model is where we build a complex neural network architecture to emulate the physical equation which captures the relationship between HSP and Hildebrand solubility. We compare the results to those from our previous work and discuss the benefits of physics-infused architectures and multi-task learning. We also use these models for transfer learning to predict HSP for heavy or high molecular weight solvents and compare our results with those in literature. All of our models are generated using our group’s machine learning and informatics program suit ChemML, where we have also created a module to display 3-dimensional Hansen spheres for to view solvent-polymer interactions.

Predicting Hildebrand and Hansen solubility parameters using multi-task learning and advanced deep neural network architectures

Macromolecular Chemistry: The Second Century

[Author Benefits](https://scimeetings.acs.org/?utm_source=pubs_content_marketing&utm_medium=website&utm_campaign=0320_MCF_NPI_Launch_Spring_Homepage&ref=pubs_content_marketing)

[How to use SciMeetings](https://storage.googleapis.com/dl.morressier.com/acs/SciMeetings-ACSSpring2020-HowToGuide_M3020-Final.pdf)

[Frequently Asked Questions](https://storage.googleapis.com/dl.morressier.com/acs/SciMtng_ACSNMSp2020_FAQ_M3120_Final.pdf)

ACS Spring 2020 National Meeting & Expo

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.