SMILES Enumeration as Data Augmentation for Neural Network Modeling of Molecules

By Esben Jannik Bjerrum. 2017

351 citations recorded by the source corpus.

Simplified Molecular Input Line Entry System (SMILES) is a single line text representation of a unique molecule. One molecule can however have multiple SMILES strings, which is a reason that canonical SMILES have been defined, which ensures a one to one correspondence between SMILES string and molecule. Here the fact that multiple SMILES represent the same molecule is explored as a technique for data augmentation of a molecular QSAR dataset modeled by a long short term memory (LSTM) cell based neural network. The augmented dataset was 130 times bigger than the original. The network trained with the augmented dataset shows better performance on a test set when compared to a model built with only one canonical SMILES string per molecule. The correlation coefficient R2 on the test set was improved from 0.56 to 0.66 when using SMILES enumeration, and the root mean square error (RMS) likewise fell from 0.62 to 0.55. The technique also works in the prediction phase. By taking the average per molecule of the predictions for the enumerated SMILES a further improvement to a correlation coefficient of 0.68 and a RMS of 0.52 was found.

Subfields: Computational Theory and Mathematics, Materials Chemistry, Molecular Biology

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computational Theory and Mathematics

In Materials Chemistry

In Molecular Biology

Citations in the corpus

Using Rule-Based Labels for Weak Supervised Learning: A ChemNet for Transferable Chemical Property Prediction (2017, methodology): “Other approaches use 2D or 3D images to train convolutional neural network (CNN) models [14, 15, 31], or SMILES strings to train recurrent neural network (RNN) models [3, 12].”
The Effectiveness of Data Augmentation in Image Classification using Deep Learning (2017, methodology): “Traditional transformations consist of using a combination of afﬁne transformations to manipulate the training data [9].”
Learning Deep Generative Models of Graphs (2018, methodology): “…for the LSTM on SMILES, we convert the SMILES into a graph, permute the F i x e d O r d e r Step 5 Step 15 Step 25 Step 35 Final Sample R a ndo m O r d e r Step 5 Step 15 Step 25 Step 35 Final Sample node ordering and then convert back to SMILES without canonicalization, similar to (Bjerrum, 2017).”
Syntax-Directed Variational Autoencoder for Structured Data (2018, background): “Techniques including data augmentation (Bjerrum, 2017), active learning (Janz et al., 2017) and reinforcement learning (Guimaraes et al., 2017) also been proposed to tackle this issue.”
SMILES Transformer: Pre-trained Molecular Fingerprint for Low Data Drug Discovery (2019, methodology): “To alleviate bias for the canonical representation of SMILES, we randomly transformed them every time they were used by the SMILES enumerator (Bjerrum 2017).”
ChemNet: A Transferable and Generalizable Deep Neural Network for Small-Molecule Property Prediction (2017, methodology): “Other approaches use 2D or 3D images to train convolutional neural network (CNN) models [9, 10, 11], or SMILES strings to train recurrent neural network (RNN) models [12].”
Molecular representations in AI-driven drug discovery: a review and practical guide (2020, methodology): “Data augmentation can be done for many applications using randomized SMILES [51, 138].”
Large Language Models on Graphs: A Comprehensive Survey (2023, background): “Numerous studies [152]–[154] have shown that training on different string-based views of the same molecule can improve the sequential model’s performance, as these data augmentation approaches manage to retain the permutation-invariance nature of graphs.”
State-of-the-art augmented NLP transformer models for direct and single-step retrosynthesis (2020, methodology): “for image recognition [40], and have been already successfully used in the context of several chemical problems [27, 28, 29, 30], including reaction predictions [18, 31], but were limited to the input data.”
Do Large Scale Molecular Language Representations Capture Important Structural Information? (2021, methodology): “Following these methods, RNN-based models have been used for molecular representation learning using SMILES and other linear molecular annotations as inputs Bjerrum (2017). More recent directions involve graph convolutional networks which encode molecular graphs into neural fingerprints. To better c”
SMILES2Vec: An Interpretable General-Purpose Deep Neural Network for Predicting Chemical Properties (2017, neutral): “In terms of text representations, we acknowledge there has been some prior work in this direction [2, 22].”
MolGenSurvey: A Systematic Survey in Machine Learning Models for Molecule Design (2022, methodology): “For molecular string representation, existing work considers advanced representation methods from other domains, including convolutional neural network (CNN) [19], recurrent neural network (RNN) [20], [21], [22], and Transformer [23], [24], [25].”