Substructure-based neural machine translation for retrosynthetic prediction

Abstract

With the rapid improvement of machine translation approaches, neural machine translation has started to play an important role in retrosynthesis planning, which finds reasonable synthetic pathways for a target molecule. Previous studies showed that utilizing the sequence-to-sequence frameworks of neural machine translation is a promising approach to tackle the retrosynthetic planning problem. In this work, we recast the retrosynthetic planning problem as a language translation problem using a template-free sequence-to-sequence model. The model is trained in an end-to-end and a fully data-driven fashion. Unlike previous models translating the SMILES strings of reactants and products, we introduced a new way of representing a chemical reaction based on molecular fragments. It is demonstrated that the new approach yields better prediction results than current state-of-the-art computational methods. The new approach resolves the major drawbacks of existing retrosynthetic methods such as generating invalid SMILES strings. Specifically, our approach predicts highly similar reactant molecules with an accuracy of 57.7%. In addition, our method yields more robust predictions than existing methods.

RetroTRAE: retrosynthetic translation of atomic environments with Transformer

“…This result surpassed all current state-of-the-art approaches by a large margin. [29], the similarity score can be considered an effective metric for assessing the retrosynthetic quality of predictions. High similarity scores indicate higher-quality retrosynthetic predictions.…”

“…The product-reactant pairs were carefully curated in the same manner as in our previous study [29]. As a result, we generated two distinct curated datasets consisting of unimolecular (P =⇒ R) and bimolecular (P =⇒ R 1 + R 2 ) reactions, with sizes 100K and 314K respectively.…”

“…In addition, grammatically valid SMILES are not guaranteed to be semantically valid or synthetically accessible. In our previous study [29], we demonstrated that representing molecules as the sets of fragments is an effective solution to the aforementioned problems.…”

RetroTRAE: retrosynthetic translation of atomic environments with Transformer

“…Also, there are no reaction class information is available in this dataset. The product-reactant pairs are carefully curated in the same manner with our previous study [29]. As a result, we generated two distinct curated datasets consist of unimolecular (P =⇒ R) and bimolecular (P =⇒ R 1 + R 2 ) reactions, with sizes 100K and 314K respectively.…”

“…As we have stressed in our previous report [29], the similarity score can be seen as an effective metric to assess the retrosynthetic quality of predictions. High similarity scores indicate higher-quality retrosynthetic predictions.…”