DeepGraphMolGen, a multi-objective, computational strategy for generating molecules with desirable properties: a graph c
- PDF / 3,171,311 Bytes
- 17 Pages / 595.276 x 790.866 pts Page_size
- 80 Downloads / 168 Views
RESEARCH ARTICLE
Journal of Cheminformatics Open Access
DeepGraphMolGen, a multi‑objective, computational strategy for generating molecules with desirable properties: a graph convolution and reinforcement learning approach Yash Khemchandani1,2, Stephen O’Hagan3, Soumitra Samanta1, Neil Swainston1, Timothy J. Roberts1, Danushka Bollegala4 and Douglas B. Kell1,5*
Abstract We address the problem of generating novel molecules with desired interaction properties as a multi-objective optimization problem. Interaction binding models are learned from binding data using graph convolution networks (GCNs). Since the experimentally obtained property scores are recognised as having potentially gross errors, we adopted a robust loss for the model. Combinations of these terms, including drug likeness and synthetic accessibility, are then optimized using reinforcement learning based on a graph convolution policy approach. Some of the molecules generated, while legitimate chemically, can have excellent drug-likeness scores but appear unusual. We provide an example based on the binding potency of small molecules to dopamine transporters. We extend our method successfully to use a multi-objective reward function, in this case for generating novel molecules that bind with dopamine transporters but not with those for norepinephrine. Our method should be generally applicable to the generation in silico of molecules with desirable properties. Keywords: Cheminformatics, Deep learning, Generative methods, QSAR, Reinforcement learning Introduction The in silico (and experimental) generation of molecules or materials with desirable properties is an area of immense current interest (e.g. [1–28]). However, difficulties in producing novel molecules by current generative methods arise because of the discrete nature of chemical space, as well as the large number of molecules [29]. For example, the number of drug-like molecules has been *Correspondence: [email protected] 1 Department of Biochemistry and Systems Biology, Institute of Systems, Molecular and Integrative Biology, University of Liverpool, Crown St, Liverpool L69 7ZB, UK Full list of author information is available at the end of the article
estimated to be between 1023 and 1060 [30–34]. Moreover, a slight change in molecular structure can lead to a drastic change in a molecular property such as binding potency (so-called activity cliffs [35–37]). Earlier approaches to understanding the relationship between molecular structure and properties used methods such as random forests [38, 39], shallow neural networks [40, 41], Support Vector Machines [42], and Genetic Programming [43]. However, with the recent developments in Deep Learning [44, 45], deep neural networks have come to the fore for property prediction tasks [3, 46–48]. Notably, Coley et al. [49] used Graph convolutional networks effectively as a feature encoder for input to the neural network.
© The Author(s) 2020. This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, shar
Data Loading...
