EvoMol: a flexible and interpretable evolutionary algorithm for unbiased de novo molecular generation

Abstract

The objective of this work is to design a molecular generator capable of exploring known as well as unfamiliar areas of the chemical space. Our method must be flexible to adapt to very different problems. Therefore, it has to be able to work with or without the influence of prior data and knowledge. Moreover, regardless of the success, it should be as interpretable as possible to allow for diagnosis and improvement. We propose here a new open source generation method using an evolutionary algorithm to sequentially build molecular graphs. It is independent of starting data and can generate totally unseen compounds. To be able to search a large part of the chemical space, we define an original set of 7 generic mutations close to the atomic level. Our method achieves excellent performances and even records on the QED, penalised logP, SAscore, CLscore as well as the set of goal-directed functions defined in GuacaMol. To demonstrate its flexibility, we tackle a very different objective issued from the organic molecular materials domain. We show that EvoMol can generate sets of optimised molecules having high energy HOMO or low energy LUMO, starting only from methane. We can also set constraints on a synthesizability score and structural features. Finally, the interpretability of EvoMol allows for the visualisation of its exploration process as a chemically relevant tree.

MolFinder: an evolutionary algorithm for the global optimization of molecular properties and the extensive exploration of chemical space using SMILES

“…4 have low values, less than 2.5, suggesting that they are readily synthesizable. It is noticeable that, even though we optimized in this study, the values of the top-12 molecules are identical or comparable to the best reported values obtained from the sole optimization of [ 26 ]. In conclusion, the above results indicate that molecule optimization of using MolFinder successfully generated a set of molecules with good drug-likeness and synthetic accessibility simultaneously.…”

“…Most existing GA-based molecular optimization algorithms are based on the graph representation of a molecule. In recent studies, they showed competitive, sometimes better, performance compared to ML-based methods in generating novel molecules with desired properties [ 26 , 27 , 29 , 30 ]. In addition, the design of any arbitrary operation may be limited because generally it is tightly coupled with the molecular manipulation functionality of underlying cheminformatics libraries, such as RDKit [ 35 ].…”

MolFinder: An Efficient Global Molecular Property Optimization and Search Algorithm Using SMILES

“…It is noticeable that, even though we optimized S mQED in this study, the S QED values of the top-12 molecules are identical or comparable to the best reported values obtained from the sole optimization of S QED . 25 In conclusion, the above results indicate that molecule optimization of S mQED using MolFinder successfully generated a set of molecules with good drug-likeness and synthetic accessibility simultaneously. This clearly demonstrates that MolFinder can help accelerate the drug discovery process by generating novel drug candidates that are readily synthesizable.…”

“…In addition to recent ML-based approaches, various genetic algorithm (GA)-based molecular property optimization algorithms have been developed. [25][26][27][28][29][30][31][32][33] The main advantage of GA-based algorithms is that they do not require a large amount of molecule data relevant to a given optimization task because they search novel molecules in a combinatorial and stochastic way. Also, they do not need to train a molecule generator, which takes considerable computational time and resources.…”

“…In recent studies, they showed competitive, sometimes better, performance compared to ML-based methods in generating novel molecules with desired properties. 25,26,28,29 A GA-based method using the graph representation requires careful design of crossover and/or mutation operations of graphs, which may bias the direction and extension of chemical space search. In addition, the design of any arbitrary operation may be limited because generally it is tightly coupled with the molecular manipulation functionality of underlying cheminformatics libraries, such as RDKit.…”