State-of-the-art augmented NLP transformer models for direct and single-step retrosynthesis (2020-03-05T00:00:00.000000Z)

TL;DR

It is demonstrated that applying augmentation techniques to the SMILE representation of target data significantly improves the quality of the reaction predictions.

Abstract

We investigated the effect of different training scenarios on predicting the (retro)synthesis of chemical compounds using text-like representation of chemical reactions (SMILES) and Natural Language Processing (NLP) neural network Transformer architecture. We showed that data augmentation, which is a powerful method used in image processing, eliminated the effect of data memorization by neural networks and improved their performance for prediction of new sequences. This effect was observed when augmentation was used simultaneously for input and the target data simultaneously. The top-5 accuracy was 84.8% for the prediction of the largest fragment (thus identifying principal transformation for classical retro-synthesis) for the USPTO-50k test dataset, and was achieved by a combination of SMILES augmentation and a beam search algorithm. The same approach provided significantly better results for the prediction of direct reactions from the single-step USPTO-MIT test set. Our model achieved 90.6% top-1 and 96.1% top-5 accuracy for its challenging mixed set and 97% top-5 accuracy for the USPTO-MIT separated set. It also significantly improved results for USPTO-full set single-step retrosynthesis for both top-1 and top-10 accuracies. The appearance frequency of the most abundantly generated SMILES was well correlated with the prediction outcome and can be used as a measure of the quality of reaction prediction. Development of algorithms to predict reactant and reagents given a target molecule is key to accelerate retrosynthesis approaches. Here the authors demonstrate that applying augmentation techniques to the SMILE representation of target data significantly improves the quality of the reaction predictions.

Authors

References50 items

QSAR without borders.

Current and Future Roles of Artificial Intelligence in Medicinal Chemistry Synthesis

A Graph to Graphs Framework for Retrosynthesis Prediction

Transformer-CNN: Swiss knife for QSAR modeling and interpretation

Predicting retrosynthetic pathways using transformer-based models and a hyper-graph exploration strategy

Automatic retrosynthetic route planning using template-free models

Data Augmentation and Pretraining for Template-Based Retrosynthetic Prediction in Computer-Aided Synthesis Planning

Retrosynthesis Prediction with Conditional Graph Logic Network

Predicting Retrosynthetic Reactions Using Self-Corrected Transformer Neural Networks

Learning to Make Generalizable and Diverse Predictions for Retrosynthesis

Augmentation Is What You Need!

A survey on Image Data Augmentation for Deep Learning

Prediction and Interpretable Visualization of Retrosynthetic Reactions Using Graph Convolutional Networks

A Transformer Model for Retrosynthesis

Graph Transformation Policy Network for Chemical Reaction Prediction

Synergy Effect between Convolutional Neural Networks and the Multiplicity of SMILES for Improvement of Molecular Prediction

Molecular Transformer: A Model for Uncertainty-Calibrated Chemical Reaction Prediction

A graph-convolutional neural network model for the prediction of chemical reactivity

Efficient Syntheses of Diverse, Medicinally Relevant Targets Planned by Computer and Executed in the Laboratory

Comprehensive organic transformations : a guide to functional group preparations

Artificial intelligence in synthetic chemistry: achievements and prospects

Computer-Assisted Retrosynthesis Based on Molecular Similarity

“Found in Translation”: predicting outcomes of complex organic chemistry reactions using neural sequence-to-sequence models† †Electronic supplementary information (ESI) available: Time-split test set and example predictions, together with attention weights, confidence and token probabilities. See DO

Predicting Organic Reaction Outcomes with Weisfeiler-Lehman Network

Planning chemical syntheses with deep neural networks and symbolic AI

Attention is All you Need

Retrosynthetic Reaction Prediction Using Neural Sequence-to-Sequence Models

Neural-Symbolic Machine Learning for Retrosynthesis and Reaction Prediction.

Prediction of Organic Reaction Outcomes Using Machine Learning

SMILES Enumeration as Data Augmentation for Neural Network Modeling of Molecules

Linking the Neural Machine Translation and the Prediction of Organic Chemistry Reactions

Computer-Assisted Synthetic Planning: The End of the Beginning.

Sequence to Sequence Learning with Neural Networks

Extraction of chemical structures and reactions from the literature

Route Designer: A Retrosynthetic Analysis Tool Utilizing Automated Retrosynthetic Rule Generation

Neural network studies, 1. Comparison of overfitting and overtraining

Neural Network Studies

End of the beginning

SMILES. 2. Algorithm for generation of unique SMILES notation

The Logic of Chemical Synthesis

SMILES, a chemical language and information system. 1. Introduction to methodology and encoding rules

Computer-assisted analysis in organic synthesis.

Workshop and Special Sessions, Münich, 17–19th September 2019

Publisher's Note

SOPHIA, a Knowledge Base-Guided Reaction Prediction System - Utilization of a Knowledge Base Derived from a Reaction Database

Achievements and Prospects

Top-5 error of the previous model for the mixed set was 100-94.2 = 5.8%. The error of our model is 100 -96.1 = 3.9%. The relative decrease of the error is