Generating molecules from scratch with bespoke properties is one of the most challenging tasks in chemistry. In the past few years, generative deep learning has remarkably enhanced the field of de novo design, by allowing the generation of molecules with desired properties on demand.

Chemical language models – which generate new molecules in the form of strings using deep learning – have been particularly successful in de novo drug design. Thanks to advances in natural language processing methods and interdisciplinary collaborations, chemical language models are expected to become increasingly relevant in drug discovery. Despite their promise, many challenges still await to be tackled, such as the efficient exploration of the vast ‘chemical universe’ in search for novel chemical matter, a fine-grained control of the properties of the molecular designs, and the capacity to generate molecules with an array of desirable features on demand. This project, spearheaded by R. Özçelik, aims to boost the potential of chemical language models in drug discovery, by tailoring deep learning innovation to the molecular world. 

Selected references

Grisoni F (2023). Chemical language models for de novo drug design: Challenges and opportunities. Current Opinion in Structural Biology 79, 102527.
doi.org/10.1016/j.sbi.2023.102527

Özçelik R, de Ruiter S, Grisoni F (2023). Structured State-Space Sequence Models for De Novo Drug Design. ChemRxiv. doi.org/10.26434/chemrxiv-2023-jwmf3