GuacaMol: Benchmarking Models for De Novo Molecular Design

Abstract

De novo design seeks to generate molecules with required property profiles by virtual design-make-test cycles. With the emergence of deep learning and neural generative models in many application areas, models for molecular design based on neural networks  appeared  recently  and  show  promising  results. However,  the  new  models have not been profiled on consistent tasks, and comparative studies to well-established algorithms have only seldom been performed.  To standardise the assessment of both classical  and  neural  models  for de novo molecular  design, we  propose  an  evaluation framework, GuacaMol, based on a suite of standardised benchmarks. The benchmark tasks encompass measuring the fidelity of the models to reproduce the property distribution of the training sets, the ability to generate novel molecules, the exploration and exploitation of chemical space, and a variety of single and multi-objective optimisation tasks.  The benchmarking framework is available as an open-source Python package.