Generating Focussed Molecule Libraries for Drug Discovery with Recurrent Neural Networks

Abstract

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.

arXiv comment17 pages, 17 figures

References in corpus (10)

• Adam: A Method for Stochastic Optimization
• LSTM: A Search Space Odyssey
• Molecular Graph Convolutions: Moving Beyond Fingerprints
• Exploring the Limits of Language Modeling
• Massively Multitask Networks for Drug Discovery
• Google's Multilingual Neural Machine Translation System: Enabling Zero-Shot Translation
• Toxicity Prediction using Deep Learning
• Learning Python Code Suggestion with a Sparse Pointer Network
• Modeling Industrial ADMET Data with Multitask Networks
• Low Data Drug Discovery with One-shot Learning

Cited by in corpus (24)

• Learning to Plan Chemical Syntheses
• Junction Tree Variational Autoencoder for Molecular Graph Generation
• Learning Deep Generative Models of Graphs
• ChemTS: An Efficient Python Library for de novo Molecular Generation
• Transformer-CNN: Fast and Reliable tool for QSAR
• Learning Multimodal Graph-to-Graph Translation for Molecular Optimization
• GraphVAE: Towards Generation of Small Graphs Using Variational Autoencoders
• Grammar Variational Autoencoder
• Hierarchical Generation of Molecular Graphs using Structural Motifs
• ChemGAN challenge for drug discovery: can AI reproduce natural chemical diversity?
• Application of generative autoencoder in de novo molecular design
• Synergy Effect between Convolutional Neural Networks and the Multiplicity of SMILES for Improvement of Molecular Prediction
• Re-evaluating Retrosynthesis Algorithms with Syntheseus
• In silico generation of novel, drug-like chemical matter using the LSTM neural network
• Hierarchical Graph-to-Graph Translation for Molecules
• Sequence Tutor: Conservative Fine-Tuning of Sequence Generation Models with KL-control
• Towards "AlphaChem": Chemical Synthesis Planning with Tree Search and Deep Neural Network Policies
• DeepCCI: End-to-end Deep Learning for Chemical-Chemical Interaction Prediction
• Learning a Continuous Representation of 3D Molecular Structures with Deep Generative Models
• Barking up the right tree: an approach to search over molecule synthesis DAGs
• Improving Molecular Design by Stochastic Iterative Target Augmentation
• "Found in Translation": Predicting Outcomes of Complex Organic Chemistry Reactions using Neural Sequence-to-Sequence Models
• Retrosynthetic reaction prediction using neural sequence-to-sequence models
• Probabilistic hypergraph grammars for efficient molecular optimization