Maxsmi: Maximizing molecular property prediction performance with confidence estimation using SMILES augmentation and deep learning

Kimber, Talia B.; Gagnebin, Maxime; Volkamer, Andrea

doi:10.1016/j.ailsci.2021.100014

Cited by 16 publications

(31 citation statements)

References 33 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Here, we explored three types of neural networks suitable to learn from SMILES strings: 1. Convolutional neural networks (CNN) 62 . This neural network architecture uses a learnable convolutional filter to aggregate information from neighboring positions in a SMILES string with a sliding window approach.…”

Section: Smiles-based Deep Learning Methodsmentioning

confidence: 99%

See 1 more Smart Citation

Exposing the limitations of molecular machine learning with activity cliffs.

Tilborg

Alenicheva²,

Grisoni

2022

Preprint

View full text Add to dashboard Cite

Machine learning has become a crucial tool in drug discovery and chemistry at large, e.g., to predict molecular properties, such as bioactivity, with high accuracy. However, activity cliffs – pairs of molecules that are highly similar in their structure but exhibit large differences in potency – have been underinvestigated for their effect on model performance. Not only are these edge cases informative for molecule discovery and optimization, but models that are well-equipped to accurately predict the potency of activity cliffs have an increased potential for prospective applications. Our work aims to fill the current knowledge gap on best-practice machine learning methods in the presence of activity cliffs. We benchmarked a total of 720 machine and deep learning models on curated bioactivity data from 30 macromolecular targets for their performance on activity cliff compounds. While all methods struggled in the presence of activity cliffs, machine learning approaches based on molecular descriptors outperformed more complex deep learning methods. Our findings highlight large case-by-case differences in performance, advocating for (a) the inclusion of dedicated “activity-cliff-centered” metrics during model development and evaluation, and (b) the development of novel algorithms to better predict the properties of activity cliffs. To this end, the methods, metrics, and results of this study have been encapsulated into an open-access benchmarking platform named MoleculeACE (Activity Cliff Estimation, available on GitHub at: https://github.com/molML/MoleculeACE). MoleculeACE is designed to steer the community towards addressing the pressing but overlooked limitation of molecular machine learning models posed by activity cliffs.

show abstract

Section: Smiles-based Deep Learning Methodsmentioning

confidence: 99%

“…2) for 100 epochs with a regression head. (b) 1D CNNs were adapted from a recent study 62 . We used a single 1D convolutional layer with a stepsize equal to 1, followed by a fully connected layer, with training for 500 epochs.…”

Section: Data Curationmentioning

confidence: 99%

Exposing the limitations of molecular machine learning with activity cliffs.

Tilborg

Alenicheva²,

Grisoni

2022

Preprint

View full text Add to dashboard Cite

show abstract

“…Some data augmentation methods used in deep learning also share similarities with features perturbation. For example, considering that SMILES (Simplified molecular input line entry system) of a molecule are not unique, Kimber et al used different SMILES to represent the same molecule for data augmentation, where SMILES are the input format of their model (Kimber et al, 2021). Similar to features perturbation, different SMILES can provide different perspectives on the same molecule.…”

Section: Features Perturbationmentioning

confidence: 99%

Uncertainty quantification: Can we trust artificial intelligence in drug discovery?

Wang

Zheng

2022

iScience

View full text Add to dashboard Cite

“…However, RNNs are not suitable to capture localized patterns such as functional motifs. Therefore, RNNs and CNNs are often used together to complement each other [99] , [128] , [129] in chemical domains.…”

Section: Deep Learning Technologies: How Well Can We Accomplish the T...mentioning

confidence: 99%

“…The substructures can be obtained by using any chemical fingerprints [91] , [130] or fragmentation algorithm [131] . A SMILES string is just one of many possible views on a chemical compound and it is possible to use multiple SMILES representations of a compound, called SMILES augmentation [35] , [128] , [129] , [132] .…”

Section: Deep Learning Technologies: How Well Can We Accomplish the T...mentioning

confidence: 99%

On modeling and utilizing chemical compound information with deep learning technologies: A task-oriented approach

Lim¹,

Lee²,

Piao³

et al. 2022

Computational and Structural Biotechnology Journal

View full text Add to dashboard Cite

Maxsmi: Maximizing molecular property prediction performance with confidence estimation using SMILES augmentation and deep learning

Cited by 16 publications

References 33 publications

Exposing the limitations of molecular machine learning with activity cliffs.

Exposing the limitations of molecular machine learning with activity cliffs.

Uncertainty quantification: Can we trust artificial intelligence in drug discovery?

On modeling and utilizing chemical compound information with deep learning technologies: A task-oriented approach

Contact Info

Product

Resources

About