Predicting glass transition temperature via machine learning
A new prediction method powered by machine learning and molecular embeddings to predict glass transition temperature of organic compounds with unprecedented accuracy.
Gas-particle partitioning of secondary organic aerosols is impacted by particle phase state and viscosity, which can be inferred from the glass transition temperature (Tg) of the constituting organic compounds.
The Challenge
Several parametrizations were developed to predict Tg of organic compounds based on molecular properties and elemental composition, but they are subject to relatively large uncertainties as they do not account for molecular structure and functionality.
Our Approach
We developed a new Tg prediction method powered by machine learning and “molecular embeddings”—unique numerical representations of chemical compounds that retain information on their structure, inter-atomic connectivity and functionality.
We trained multiple state-of-the-art machine learning models on databases of experimental Tg of organic compounds and their corresponding molecular embeddings.
Results
The best prediction model is the tgBoost model built with an Extreme Gradient Boosting (XGBoost) regressor trained via a nested cross-validation method, reproducing experimental data very well with a mean absolute error of 18.3 K.
The model can quantify the influence of number and location of functional groups on the Tg of organic molecules, while accounting for atom connectivity and predicting different Tg for compositional isomers.
Functional Group Sensitivity
The tgBoost model suggests the following trend for sensitivity of Tg to functional group addition:
–COOH (carboxylic acid) > –C(=O)OR (ester) ~ –OH (alcohol) > –C(=O)R (ketone) ~ –COR (ether) ~ –C(=O)H (aldehyde)
Melting Point Prediction
We also developed a model to predict the melting point (Tm) of organic compounds by training a deep neural network on a large dataset of experimental Tm. The model performs reasonably well with a mean absolute error of 31.0 K.
These new machine learning powered models can be applied to field and laboratory measurements as well as atmospheric aerosol models to predict the Tg and Tm of SOA compounds for evaluation of the phase state and viscosity of SOA.
Published in Environmental Science: Atmospheres Read the full paper →