Peptide Resources

THPdb is a manually curated database of FDA-approved therapeutic peptides and proteins, integrating 985 research publications, 70 patents, and resources like DrugBank. Current version holds 852 entries covering 239 FDA-approved therapeutics and 380 drug variants, providing comprehensive info on sequences, chemical properties, indications, mechanisms, administration routes, etc., with structural annotations for most entries. User-friendly tools, web interface, and mobile app support drug research.

TriStack is an interpretable model for precise identification of antimicrobial peptides (AMPs) and anti-inflammatory peptides (AIPs), achieving efficient prediction through three-tier feature fusion and residual network stacking. The model first extracts three types of functional features (composition, distribution, physicochemical properties) from peptide sequences, then employs a two-module architecture: the first module integrates preliminary predictions from three machine learning algorithms, and the second refines results via multi-layer residual neural network (ResNet).

xDeep-AcPEP is a CNN-based deep learning tool for predicting biological activities (EC50, LC50, IC50, LD50) of anticancer peptides (ACPs) against six tumor cell types (breast, colon, etc.).

YADAMP is a literature-curated database specializing in antimicrobial peptides (AMPs) active against bacteria, containing 2,133 peptides with detailed antibacterial activity data. Distinct from other AMP databases, it explicitly annotates activity against common bacterial strains and supports complex web-based queries (e.g., by peptide name, amino acid count, charge, hydrophobicity, sequence motifs, structure, and antibacterial activity). Designed for rapid information retrieval and structure-function analysis, it offers intuitive data visualization and statistical tools, supporting AMP mechanism research and drug development.

An ensemble learning model for anti-angiogenic peptide (AAPs) prediction, achieving precise identification by fusing classifiers with high sensitivity and specificity. The model first uses Bi-profile Bayes (BpB) features to construct the feature space, combines Relief algorithm and Incremental Feature Selection (IFS) for discriminative feature screening, and optimizes classification performance via ensemble strategy.