Conotoxins are disulfide-rich therapeutic peptides with high affinity and selectivity for ion channels, yet their optimization is hindered by limited sequence diversity and laborious engineering. Here, the authors introduce CreoPep, a deep learning-based generative framework that integrates a progressive masking strategy and an augmentation pipeline that combines physics-based energy screening with temperature-controlled multinomial sampling, rationally designing and generating diverse and potent conotoxin variants.
- Cheng Ge
- Han-Shen Tae
- Rilei Yu