Advancing Vapor Pressure Prediction: A Machine Learning Approach with Directed Message Passing Neural Networks

Abstract

Background: Vapor pressure is a critical property in chemical and environmental engineering. Accurately predicting vapor pressure across a range of temperatures is vital for various applications, but traditional methods rely on critical property measurements or quantum mechanical calculations, which can be limiting, especially for new or under-characterized chemicals. Methods: This study employs a machine learning model based on the directed message passing neural network (D-MPNN) architecture to predict the vapor pressure of organic molecules. Various strategies to incorporate temperature effects into the model are explored to improve prediction accuracy. Significant findings: The D-MPNN model achieves significantly better accuracy than the traditional PR + COSMOSAC method, with a lower average absolute relative deviation (AARD) of 0.617 compared to 1.36 for the traditional method, using a dataset of 19,079 molecules. The machine learning approach offers a robust alternative that does not require additional critical property data or quantum mechanical calculations.