Investigation of chemical structure recognition by encoder-decoder models in learning progress

Descriptor generation methods using latent representations of Encoder-Decoder (ED) models with SMILES as input is useful because of continuity of descriptor and restorability to structure. However, it is not clear how the structure is recognized in the learning progress of ED model. In this work, we created ED models of various learning progress and investigated the relationship between structural information and the learning progress. We showed that compound substructures were learned early in ED models by monitoring the accuracy of downstream tasks and input-output substructure similarity using substructure-based descriptor, which suggests that existing evaluation methods based on the accuracy of downstream tasks may not be sensitive enough to evaluate the performance of ED models few with SMILES as descriptor generation methods. On the other hand, we showed that structure restoration was time consuming, and in particular, insufficient learning led to estimation of a larger structure than the actual one. It can be inferred that determining the end point of the structure is a difficult task for the model. To the best of our knowledge, this is the first study to link the learning progress of SMILES by ED model to chemical structures for a wide range of chemicals.