Independent Vector Analysis for Molecular Data Fusion: Application to Property Prediction and Knowledge Discovery of Energetic Materials

Due to its high computational speed and accuracy compared to ab-initio quantum chemistry and forcefield modeling, the prediction of molecular properties using machine learning has received great attention in the fields of materials design and drug discovery. A main ingredient required for machine learning is a training dataset consisting of molecular features—for example fingerprint bits, chemical descriptors, etc. that adequately characterize the corresponding molecules. However, choosing features for any application is highly non-trivial, since no "universal" method for feature selection exists. In this work, we propose a data fusion framework that uses Independent Vector Analysis to uncover underlying complementary information contained in different molecular featurization methods. Our approach takes an arbitrary number of individual feature vectors and generates a low dimensional set of features—molecular signatures—that can be used for the prediction of molecular properties and for knowledge discovery. We demonstrate this on a small and diverse dataset consisting of energetic compounds for the prediction of several energetic properties as well as for demonstrating how to provide insights onto the relationships between molecular structures and properties.