Low Data Drug Discovery with One-Shot Learning
By Han Altae-Tran, Bharath Ramsundar, Aneesh S. Pappu, Aneesh S. Pappu, Vijay S. Pande. 2016
773 citations recorded by the source corpus.
Recent advances in machine learning have made significant contributions to drug discovery. Deep neural networks in particular have been demonstrated to provide significant boosts in predictive power when inferring the properties and activities of small-molecule compounds (Ma, J. et al. J. Chem. Inf. Model.2015, 55, 263–2742563532410.1021/ci500747n). However, the applicability of these techniques has been limited by the requirement for large amounts of training data. In this work, we demonstrate how one-shot learning can be used to significantly lower the amounts of data required to make meaningful predictions in drug discovery applications. We introduce a new architecture, the iterative refinement long short-term memory, that, when combined with graph convolutional neural networks, significantly improves learning of meaningful distance metrics over small-molecules. We open source all models introduced in this work as part of DeepChem, an open-source framework for deep-learning in drug discovery (Ramsundar, B. deepchem.io. https://github.com/deepchem/deepchem, 2016).
Subfields: Computational Theory and Mathematics, Materials Chemistry, Molecular Biology, Meta & Continual Learning, Computational Chemistry & Drug Discovery
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
- Gisbert Schneider
- Francesca Grisoni
- Chaopeng Shen
- Ola Engkvist
- José Jiménez-Luna
- Hongming Chen
- Shion Honda
- Shoi Shi
In Materials Chemistry
- Gisbert Schneider
- Francesca Grisoni
- Ola Engkvist
- Chaopeng Shen
- Hongming Chen
- Chi Chen
- Marcus Olivecrona
- Thomas Blaschke
In Molecular Biology
- Chaopeng Shen
- Ola Engkvist
- Francesca Grisoni
- Fei Wang
- Gisbert Schneider
- Hongming Chen
- Shion Honda
- Shoi Shi
In Meta & Continual Learning
- Luca Bertinetto
- Steinar Laenen
- Amos J. Storkey
- Paul Micaelli
- Chaopeng Shen
- Timothy M. Hospedales
- Antreas Antoniou
- Philip H. S. Torr
In Computational Chemistry & Drug Discovery
- Chaopeng Shen
- Gisbert Schneider
- Francesca Grisoni
- Shyue Ping Ong
- Chi Chen
- Junzhou Huang
- Vijay S. Pande
- Shion Honda
Citations in the corpus
- Meta-learning with differentiable closed-form solvers (2018, background): “Labels can be costly to acquire: in drug discovery, for instance, campaign budgets often limits researchers to only operate with a small amount of biological data that can be used to form predictions about properties and activities of compounds (Altae-Tran et al., 2017).”
- Unraveling Meta-Learning: Understanding Feature Representations for Few-Shot Tasks (2020, background): “When data is expensive or time consuming to obtain, training from scratch may be cost prohibitive (Altae-Tran et al., 2017).”
- Learning Graph-Level Representation for Drug Discovery (2017, methodology): “2015), similar graph networks that take the original molecule as input(Altae-Tran et al. 2017; Yao et al. 2017; Schütt et al. 2017; Gomes et al. 2017), and other general graph neural networks designed for all types of graph(Li et al.”
- Opportunities and obstacles for deep learning in biology and medicine (2017, methodology): “Other studies have shown the limitations of transfer learning when the prediction tasks are insufficiently related [296,311].”
- Graph Networks as a Universal Machine Learning Framework for Molecules and Crystals (2018, background): “Yet another approach to address data limitations is transfer learning 63,64 , and we have demonstrated an instructive example of how this can be applied in the case of the crystal MEGNet models.”
- Meta-Learning in Neural Networks: A Survey (2020, methodology): “In [5] an LSTM is combined with a graph neural”
- Generalizing from a Few Examples (2019, methodology): “resLSTM [3] GCN, then LSTM with attention GCN, then LSTM with attention yes cosine similarity combined”
- Meta-learning via Language Model In-context Tuning (2021, background): “intelligence, and can also relieve the burden of collecting large-scale labeled data (Altae-Tran et al., 2016; Li et al., 2019).”
- SMILES Transformer: Pre-trained Molecular Fingerprint for Low Data Drug Discovery (2019, methodology): “In these studies, the models are designed for large fully-labeled training data settings, which requires huge labeled datasets and a QSPR model for one-shot learning (Altae-Tran et al. 2017).”
- Converging blockchain and next-generation artificial intelligence technologies to decentralize and accelerate biomedical research and healthcare (2015, background): “Medical chemistry is one of the fields where data is scarce, therefore, to address this problem Altae-Tran and colleagues proposed a one-shot learning approach for the prediction of molecule toxic potential [51].”
- Retrosynthetic Reaction Prediction Using Neural Sequence-to-Sequence Models (2017, methodology): “Furthermore, an interesting extension to the seq2seq models would be to use one-shot techniques 56–58 to allow our retrosynthetic reaction prediction models to make reasonable predictions for reaction classes where only a few example reactions are available for training.”