Macrocyclization of linear molecules by deep learning to facilitate macrocyclic drug candidates discovery
Problem
Generating macrocyclic analogues of bioactive acyclic molecules is difficult due to vast chemical space.
Why it matters
Macrocycles can yield novel scaffolds with improved biological activity and physicochemical properties.
What the AI did
A Transformer-based model (Macformer) learns acyclic-to-macrocyclic SMILES relationships and generates diverse, structurally novel macrocycles with compatible linkers.
Results
'Excellent performance and generalisability' on internal ChEMBL and external ZINC test sets; no numbers stated
Impact
Evidence supports generation of diverse novel macrocycles validated on external datasets.
Record
| Venue | Nature Communications |
|---|---|
| Published | 2023 |
| Work type | article |
| Authors | Yanyan Diao, Dandan Liu, Huan Ge, Rongrong Zhang, Kexin Jiang, R. BAO, Xiaoqian Zhu, Hongjie Bi et al. |
| Citations | 56 (OpenAlex, at import time) |
| References | 68 |
| Model or method | Macformer (Transformer architecture) |
| Organization | not stated in source |
| Primary source | https://doi.org/10.1038/s41467-023-40219-8 |
| Open access | gold |
| Keywords | deep learning, Transformer, macrocycles, SMILES, generative model, drug discovery |
Importance rating
real but narrower advance
Macrocyclization tool for generating novel scaffolds in drug discovery; a real but narrower methodological advance, with utility shown on JAK2 inhibitor design only.