Generating Focussed Molecule Libraries for Drug Discovery with Recurrent Neural Networks

By Marwin H. S. Segler, Thierry Kogej, Christian Tyrchan, Mark P. Waller. 2017

98 citations recorded by the source corpus.

In de novo drug design, computational strategies are used to generate novel molecules with good affinity to the desired biological target. In this work, we show that recurrent neural networks can be trained as generative models for molecular structures, similar to statistical language models in natural language processing. We demonstrate that the properties of the generated molecules correlate very well with the properties of the molecules used to train the model. In order to enrich libraries with molecules active towards a given biological target, we propose to fine-tune the model with small sets of molecules, which are known to be active against that target. Against Staphylococcus aureus, the model reproduced 14% of 6051 hold-out test molecules that medicinal chemists designed, whereas against Plasmodium falciparum (Malaria) it reproduced 28% of 1240 test molecules. When coupled with a scoring function, our model can perform the complete de novo drug design cycle to generate large sets of novel molecules for drug discovery.

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

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

In Generative Models

Citations in the corpus

Sequence Tutor: Conservative Fine-Tuning of Sequence Generation Models with KL-control (2016, background): “Segler et al. (2017) demonstrated that an LSTM trained on sets of biologically active molecules can be used to generate novel molecules with similar properties.”
Planning chemical syntheses with deep neural networks and symbolic AI (2017, methodology): “[8–11] To perform CASP, the knowledge that humans “simply learn” has to be transferred into an executable program .”
Opportunities and obstacles for deep learning in biology and medicine (2017, methodology): “As deep learning models that directly output (molecular) graphs remain underexplored, generative neural networks for drug design typically represent chemicals with the simplified molecular-input line-entry system (SMILES), a standard string-based representation with characters that represent atoms, ”
Learning Multimodal Graph-to-Graph Translation for Molecular Optimization (2018, neutral): “…Prior work on molecular optimization approached the graph translation task through generative modeling (G´omez-Bombarelli et al., 2016; Segler et al., 2017; Kusner et al., 2017; Dai et al., 2018; Jin et al., 2018; Samanta et al., 2018; Li et al., 2018a) and reinforcement learning…”
Grammar Variational Autoencoder (2017, background): “However, both Segler et al. (2017) and Gómez-Bombarelli et al. (2016b) use character-level models for molecules.”
GraphRNN: Generating Realistic Graphs with Deep Auto-regressive Models (2018, methodology): “Most previous work on discovering molecule structures make use of a expert-crafted sequence representations of molecular graph structures (SMILES) (Olivecrona et al., 2017; Segler et al., 2017; G ´ omez-Bombarelli et al., 2016).”
Learning Deep Generative Models of Graphs (2018, methodology): “In addition, we also generated 100,000 samples from each model and evaluate the percentage of well-formatted molecule representations and the percentage of unique novel samples not already seen in the training set following (Segler et al., 2017; Olivecrona et al., 2017).”
Molecular de-novo design through deep reinforcement learning (2017, methodology): “However, we would like our new policy to be anchored to the prior policy, which has learnt both the syntax of SMILES and distribution of molecular structure in ChEMBL [13].”
Retrosynthetic Reaction Prediction Using Neural Sequence-to-Sequence Models (2017, background): “39 44 Recently, Nam and Kim 45 described a neural sequence-to-sequence (seq2seq) model for the forward reaction prediction task.”
Multi-objective de novo drug design with conditional graph generative model (2018, methodology): “Currently, the most popular solution is to fine-tune the existing model so that it can be suited for a specific task[12, 13, 16].”
Deep reinforcement learning for de novo drug design (2017, background): “Another very recent approach for generating focused molecule libraries with the desired bioactivity using Recurrent Neural Networks was proposed as well (30).”
Application of Generative Autoencoder in De Novo Molecular Design (2017, background): “Segler et al. [2] and Yuan et al. [3] applied a recurrent neural network (RNN) to a large number of chemical structures represented as SMILES strings to obtain generative models which learn the probability distribution of characters in a SMILES string.”
Generative Recurrent Networks for De Novo Drug Design (2017, background): “Importantly, in contrast to other related RNN models,([10,11,13]) ours does not rely on an explicit but limited SMILES vocabulary, which renders this new approach theoretically unlimited with regard to the chemical diversity of the training data.”