We are thrilled to announce the latest publication from our lab, titled "Advancing Vapor Pressure Prediction: A Machine Learning Approach with Directed Message Passing Neural Networks" in the Journal of the Taiwan Institute of Chemical Engineers. This study highlights an innovative application of machine learning to predict vapor pressure, a critical property in chemical and environmental engineering.
In this work, we developed a model based on Directed Message Passing Neural Networks (D-MPNNs) to accurately predict the vapor pressure of organic molecules across a range of temperatures. Traditional methods often rely on critical property data or quantum mechanical calculations, limiting their applicability to novel or under-characterized chemicals. Our machine learning model addresses these challenges by utilizing molecular structures and temperature data without requiring additional experimental inputs.
Key Findings:
- Enhanced Prediction Accuracy: The D-MPNN model achieved a significantly lower Average Absolute Relative Deviation (AARD) of 0.617 compared to 1.36 for traditional PR + COSMOSAC methods.
- Innovative Methodology: The model explores novel strategies like sum-pooling and mean-pooling molecular fingerprints and introduces temperature effects directly or through empirical equations to enhance accuracy.
- Broad Applicability: The approach demonstrates robust performance even in data-limited scenarios, paving the way for predicting vapor pressures of unexplored chemical species.
This research underscores the transformative potential of machine learning in molecular property prediction, offering a data-efficient framework for tackling complex chemical challenges.
For more details, access the full publication here.