NS-Pep: De novo Peptide Design with Non-Standard Amino Acids

NS-Pep is a unified flow-matching framework for co-designing peptide sequences and structures that incorporate non-standard amino acids (NSAAs). It introduces Residue Frequency-Guided Modification (RFGM), Progressive Side-chain Perception (PSP), and Interaction-Aware Weighting (IAW) to tackle the long-tailed NSAA distribution and to capture fine-grained side-chain geometry.

Peptide Generation on the General Test Set

Table 1 of the paper: comparison for peptide generation on the General test set.

Method	AAR % ↑	AAR(S) % ↑	RMSD Å ↓	SSR % ↑	BSR % ↑	AFF % ↑	scRMSD Å ↓	Diversity ↑
RFdiffusion	8.66	8.69	17.57	71.75	6.52	12.06	19.64	0.41
ProteinGenerator	7.08	7.12	17.26	75.30	4.28	8.91	23.02	0.39
PPFLOW	8.63	8.87	9.45	0.11	75.14	6.53	15.77	0.66
DiffPepBuilder	6.71	6.77	8.36	75.49	85.86	12.11	14.65	0.51
PepGLAD	6.82	6.90	7.78	78.72	81.48	10.36	13.43	0.72
PepFlow	16.30	16.76	4.98	85.80	81.50	11.60	12.32	0.50
PepFlow*	15.88	15.88	4.94	83.40	84.36	11.39	11.67	0.51
NS-Pep (ours)	22.53	22.35	4.92	85.00	81.62	17.23	11.50	0.49

Bold = best; italic = second best.

Model weights, training/inference code, and related files will be released later.

Citation

@article{guo2025ns,
  title={NS-Pep: De novo Peptide Design with Non-Standard Amino Acids},
  author={Guo, Tao and Yin, Junbo and Wang, Yu and Gao, Xin},
  journal={arXiv preprint arXiv:2510.03326},
  year={2025}
}