Automatic design with machine learning and molecular simulations has shown remarkable ability to generate new and promising drug candidates. Current models, however, still have problems in simulation concurrency and molecular diversity. Most methods generate one molecule at a time and do not allow multiple simulators to run simultaneously. Also, better molecular diversity would help to increase the chance of survival in the subsequent drug discovery process. Since many methods seek for a single optimal molecule, the diversity of molecule is not concerned.<br/><br/>We propose a new population-based approach using grammatical evolution named ChemGE. In our method, a large population of molecules is updated concurrently and evaluated by multiple simulators in parallel. Each molecule is doubly encoded as a sequence of integers called a chromosome and a SMILES string, which is generated from chromosome through a mapping process. The mapping process makes use of the context-free grammar of SMILES to yield a grammatically correct string. New molecules are generated by changing an integer in a chromosome. By repeatedly selecting good molecules from the original population and newly generated molecules, we finally obtain a set of desired molecules. We can conduct an evaluation of generated molecules simultaneously. Thanks to this inherent concurrency, we can easily expand our method to a parallel environment. The mutation process of our method creates built-in tendency to pursue diversity in the generated molecules. We conducted two types of experiments to generate 1) druglike molecules, and 2) high scoring ligands for a protein. In the first experiment, we confirmed that ChemGE outperformed other methods in score and speed of generating molecules. In docking experiments with thymidine kinase, ChemGE succeeded in generating hundreds of high-affinity molecules. Their diversity is better than that of known binding molecules in DUD-E, which increase the chance of survival in the following drug-design procedure. The generated high scoring molecules are different from the known inhibitors, which is beneficial in patentability.

Population-based <i>de novo</i> molecule generation using grammatical evolution

256th National Meeting (2018)

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.

Population-based de novo molecule generation using grammatical evolution