O07_01

SynthFormer: A Customizable Framework for Virtual Synthesis-Based Molecule Generation

Joshua OWOYEMI *, Tasuku ISHIDA

Elix, Inc.
( * E-mail: joshua.owoyemi@elix-inc.com)

Generating new molecules with specific properties is a crucial aspect of drug discovery and materials science. Traditional methods, such as virtual screening, are limited in their ability to efficiently explore the vast chemical space. Recent advances [1] in machine learning have led to the development of generative models for molecule design, but these often overlook the crucial aspect of synthesizability, hindering their practical application. We present SynthFormer, a novel framework for molecule generation and optimization that addresses synthesizability by directly incorporating chemical reactions and building blocks into the design process. This approach ensures that the generated molecules are not only novel and possess desirable properties but are also synthetically accessible, bridging the gap between computational design and experimental realization. The framework's adaptability allows for customization based on specific requirements, including the choice of chemical reactions, building blocks, and optimization algorithms and objectives, making it a valuable tool for various domains within chemistry and materials science. We evaluate the framework by exploring the de novo generation of selected patented compounds and show that the framework is able to suggest similar compounds while proposing feasible synthetic routes to achieve the generated molecules. We also compare the performance of multiple optimization approaches such as Monte Carlo Tree Search [2] and Reinforcement Learning [3] while utilizing the framework for the rediscovery of known compounds.

[1] Xiangru Tang, Howard Dai, Elizabeth Knight, Fang Wu, Yunyang Li, Tianxiao Li, Mark Gerstein. 2024. A survey of generative AI for de novo drug design: new frontiers in molecule and protein generation. Briefings in bioinformatics, 25(4), bbae338. https://doi.org/10.1093/bib/bbae338
[2] Shoichi Ishida, Tanuj Aasawat, Masato Sumita, Michio Katouda, Tatsuya Yoshizawa, Kazuki Yoshizoe, Koji Tsuda, Kei Terayama. ChemTSv2: Functional molecular design using de novo molecule generator. 2023. WIREs Comput Mol Sci.; 13(6):e1680. https://doi.org/10.1002/wcms.1680
[3] Sai Krishna Gottipati, Boris Sattarov, Sufeng Niu, Yashaswi Pathak, Haoran Wei, Shengchao Liu, Karam J. Thomas, Simon Blackburn, Connor W. Coley, Jian Tang, Sarath Chandar, and Yoshua Bengio. 2020. Learning to navigate the synthetically accessible chemical space using reinforcement learning. In Proceedings of the 37th International Conference on Machine Learning (ICML'20), Vol. 119. JMLR.org, Article 344, 3668–3679.