SN-PepDB: a deep learning-driven database for characterizing Sipunculus nudus bioactive peptides and their structure-activity relationships
Related Experiment Video
Updated: Nov 5, 2026
A Bilingual Computational Workflow for Identifying Potential PLK1 Inhibitors in American Sign Language and English
Published on: April 3, 2026
Yi Zhang1, Guandi Qin2, Le Jiang2
1Guangdong Provincial Key Laboratory of Aquatic Product Processing and Safety, Guangdong Provincial Engineering Laboratory for Marine Biological Products, Guangdong Provincial Center for Modern Agricultural Scientific Innovation, Shenzhen Institute of Guangdong Ocean University, Zhanjiang Municipal Key Laboratory of Marine Drugs and Nutrition for Brain Health, Research Institute for Marine Drugs and Nutrition, College of Food Science and Technology, Guangdong Ocean University, Zhanjiang 524088, China; Southern Marine Science and Engineering Guangdong Laboratory (Zhanjiang), Zhanjiang 524088, China.
Abstract:
Marine invertebrate peptides remain underrepresented in bioactivity databases. Their translational potential is constrained by the absence of species-specific repositories, limited integration of safety profiling, and insufficient exploitation of prediction probability distributions for threshold optimization. To address these gaps, we developed SN-PepDB, a comprehensive multi-target peptide database for Sipunculus nudus. Combinatorial enzymatic hydrolysis across 22 conditions and high-resolution LC-MS/MS yielded 10,827 unique peptides. Safety-first screening eliminated sequences predicted as potentially toxic or allergenic, securing 5463 candidates. Using the ESM-2-embedded UniDL4BioPep framework, we introduced a distribution-calibrated threshold strategy that categorizes prediction distributions into three morphological types: bimodal (threshold ≥0.90), mound-shaped (90th percentile; antioxidant ≥0.89, BBB ≥ 0.80), and right-skewed (threshold = 0.50). Compared with the conventional 0.5 cutoff, this approach concentrated candidates to 9.7-26.0% of the safe peptide population across five bioactivities: antihypertensive (n = 1422), antibacterial (n = 653), antioxidant (n = 572), BBB-permeable (n = 529), and cell-penetrating (n = 697) peptides. Comparative structure-activity relationship analysis revealed activity-specific determinants, including Pro-dominated C-terminal motifs associated with ACE inhibition. Experimental validation provided preliminary support for the prediction-prioritization workflow: the selected peptide PPVPHP showed ACE-inhibitory activity with an IC50 of 146.04 ± 4.63 μM, while the antioxidant-positive peptide LLW showed greater DPPH radical-scavenging activity than the below-threshold peptide FKVFY at 200 μM (43.1% vs. 15.5%). Molecular docking and in silico gastrointestinal digestion further supported PPVPHP as a candidate with favorable ACE-binding and predicted proteolytic-stability profiles. The interactive database is publicly accessible at https://zai4s.shinyapps.io/SN-PepDB/.
