Searching The Chemical Space For Effective Magnesium Dissolution Modulators: A Deep Learning Approach Using Sparse Features

This study aimed to predict and test the corrosion inhibition efficiencies of 15 previously unseen compounds using the ExChem21 routine. The molecules in the dataset were represented in form of a 2D map following a dimension reduction approach, thereby visualising the relationships between molecular structure and corrosion inhibition performance. The predictive performance of the neural networks was evaluated and outliers were discussed with respect to their chemical features. Furthermore, the effect of integrating more data into the training set was assessed and the scalability of the approach was confirmed.