Peptide Resources

STADIP is a stacking-based ensemble predictor for antidiabetic peptides (ADPs), enabling precise identification through multi-dimensional feature fusion and two-step feature selection. The model integrates 12 feature encoding methods (amino acid composition, physicochemical properties, sequence motifs) with 7 machine learning algorithms (XGBoost, SVM, etc.) to build 84 baseline models, employs eXtreme Gradient Boosting with Sequential Forward Selection (XGB-SFS) for optimal feature subset selection, and finally integrates 45 key features into an XGBoost classifier via a meta-predictor.

Five sets of non-trained supervised multi-query similarity search models integrated into the StarPep toolbox, designed to address the resource-intensive challenges of experimental antiviral peptide (AVPs) discovery. The models enable efficient screening by comparing query sequences with multi-dimensional similarity features (e.g., sequence motifs, physicochemical property distributions) of known AVPs.

StraPep is a structural database for bioactive peptides, housing 3,791 peptide structures (1,312 unique sequences), with 68% containing disulfide bonds (significantly higher than PDB's 21%) and 24% featuring ≥3-bond cystine knots, offering stable scaffolds for drug design. It provides detailed annotations (experimental structures, disulfide positions, secondary structures) and supports sequence/structure-based searching and visualization, facilitating peptide drug development.

TargetAntiAngio is a sequence-based tool for predicting and analyzing anti-angiogenic peptides, integrating multi-class peptide features (amino acid composition, physicochemical properties, structural features) via random forest classifier, achieving 77.50% average accuracy in independent validation—outperforming existing methods.

TargetCPP is a computational tool for accurate prediction of cell-penetrating peptides, leveraging optimized multi-scale features (including composite sequence representation, composition transition/distribution) and gradient boosting decision tree for classification.