MPGVAE: improved generation of small organic molecules using message passing neural nets

Abstract Graph generation is an extremely important task, as graphs are found throughout different areas of science and engineering. In this work, we focus on the modern equivalent Erdos–Rényi random graph model: variational autoencoder (GVAE) (Simonovsky Komodakis 2018 Int. Conf. Artificial Neural Networks pp 412–22). This model assumes edges nodes independent in order to generate entire at a time using multi-layer perceptron decoder. As result these assumptions, GVAE has difficulty matching training distribution relies expensive procedure. We improve class models by building message passing neural network into GVAE’s encoder demonstrate our specific task generating small organic molecules.