Graphs are widespread mathematical structures that can be used to describe an assortment of relational information, and are natural choices for describing molecular structures. Recently, there has been an increase in the use of graph neural networks (GNNs) for modeling patterns in graph-structured data, including graph-based molecular generation for pharmaceutical drug discovery. The guiding principle behind graph-based molecular design can be boiled down to generating graphs which meet all the criteria of desirable drug-like molecules.
We have applied GNNs to the task of molecular generation and recently published GraphINVENT[1,2], a platform for graph-based molecular design using message passing neural networks (MPNNs) and a tiered feed-forward network structure to probabilistically generate new molecules one atom/bond at a time. Graphs as data structures are very memory-intensive objects compared to the alternative, string-based approaches for molecular generation. As such, GraphINVENT was optimized so as to successfully train on large molecular datasets (e.g. millions of small molecules) using GPUs.
GraphINVENT models can quickly learn the underlying distribution of properties in training set molecules without any explicit writing of chemical rules. The proposed models perform well for molecular generative tasks when benchmarked using the Molecular Sets (MOSES) platform . Our work illustrates how deep learning methods can enhance drug design and shows that graph-based generative models merit further exploration for molecular graph generation.
1. Mercado R, Rastemo T, Lindelöf E, et al. Graph Networks for Molecular Design. 2020. doi:10.26434/chemrxiv.12843137.v1 2. Mercado R, Rastemo T, Lindelöf E, et al. Practical Notes on Building Molecular Graph Generative Models. 2020. doi:10.26434/chemrxiv.12888383.v1 3. Polykovskiy D, Zhebrak A, Sanchez-Lengeling B, et al. Molecular Sets (MOSES): A Benchmarking Platform for Molecular Generation Models. 2018. doi:arXiv:1811.12823v1