Transformer-Based Molecular Generative Model for Antiviral Drug Design

• DOI: 10.1021/acs.jcim.3c00536
• Publication Date: 2023-06-27T13:17:24Z
• Authors: Jiashun Mao; Jianmin Wang; Amir Zeb; Kwang-Hwi Cho; Haiyan Jin; Jongwan Kim; Onju Lee; Yunyun Wang; Kyoung Tai No
• Keywords: Chemical space

Since the Simplified Molecular Input Line Entry System (SMILES) is oriented to atomic-level representation of molecules and not friendly in terms human readability editable, however, IUPAC closest natural language very human-oriented performing molecular editing, we can manipulate generate corresponding new produce programming-friendly forms SMILES. In addition, antiviral drug design, especially analogue-based also more appropriate edit design directly from functional group level than atomic SMILES, since designing analogues involves altering R only, which closer knowledge-based a chemist. Herein, present novel data-driven self-supervised pretraining generative model called "TransAntivirus" make select-and-replace edits convert organic into desired properties for candidate analogues. The results indicated that TransAntivirus significantly superior control models novelty, validity, uniqueness, diversity. showed excellent performance optimization nucleoside non-nucleoside by chemical space analysis property prediction analysis. Furthermore, validate applicability drugs, conducted two case studies on screened four lead compounds against anticoronavirus disease (COVID-19). Finally, recommend this framework accelerating discovery.