RetroTRAE: retrosynthetic translation of atomic environments with Transformer

04 August 2021, Version 1
This content is a preprint and has not undergone peer review at the time of posting.

Abstract

We present the new retrosynthesis prediction method RetroTRAE using fragment-based tokenization combined with the Transformer architecture. RetroTRAE represents chemical reactions by using the changes of fragment sets of molecules using the atomic environment fragmentation scheme. Atom environments stand as an ideal, chemically meaningful building blocks together producing a high resolution molecular representation. Describing a molecule with a set of atom environments establishes a clear relationship between translated product-reactant pairs due to conservation of atoms in reactions. Our model achieved a top-1 accuracy of 67.1% within the bioactively similar range for USPTO test dataset, outperforming the other state of the art, translation methods. We investigated the effect of different encoding scenarios on predicting the reactant candidates. We also critically assessed the retrieval process that converts a set of fragments into a molecule with respect to coverage, degeneracy and resolution. Our new template-free model for retrosynthetic prediction provides fast and reliable retrosynthetic route planning for substances whose fragmentation patterns are revealed.

Keywords

Retrosynthesis
Transformer
Neural Machin Translation
Atom Environments

Comments

Comments are not moderated before they are posted, but they can be removed by the site moderators if they are found to be in contravention of our Commenting Policy [opens in a new tab] - please read this policy before you post. Comments should be used for scholarly discussion of the content in question. You can find more information about how to use the commenting feature here [opens in a new tab] .
This site is protected by reCAPTCHA and the Google Privacy Policy [opens in a new tab] and Terms of Service [opens in a new tab] apply.