Peptide Resources

PepNet is an interpretable neural network leveraging pre-trained protein language models to predict both anti-inflammatory peptides (AIPs) and antimicrobial peptides (AMPs). By integrating residual dilated convolution and Transformer modules, it fully exploits residue arrangement and physicochemical properties from sequences. Outperforming state-of-the-art tools in independent tests, PepNet visualizes attention weights to highlight functionally critical regions, enhancing interpretability. The user-friendly web server enables batch prediction and result visualization, providing an efficient platform for immunotherapy and antimicrobial drug development.

PEPred-Suite is the first bioinformatics tool enabling generic prediction of 8 types of therapeutic peptides (e.g., antimicrobial, anti-inflammatory) via adaptive feature representation. It employs a two-step optimization: integrating multi-dimensional descriptors (amino acid composition, sequence order) and selecting discriminative features via AUC-based ranking, followed by training random forest models for each peptide type.

PEPred-Suite is a bioinformatics tool for generic prediction of therapeutic peptides, enabling high-precision identification across multiple peptide types via adaptive feature representation. It introduces a two-step feature optimization strategy: integrating diverse sequence-based descriptors (e.g., amino acid composition, sequence order features), incorporating learned class information into features, and extracting discriminative features based on the area under the receiver operating characteristic curve (AUC).

PhytAMP, which contains valuable information on antimicrobial plant peptides, including taxonomic, microbiological and physicochemical data. Information is very easy to extract from this database and allows rapid prediction of structure/function relationships and target organisms and hence better exploitation of plant peptide biological activities in both the pharmaceutical and agricultural sectors.

PIP-EL is a novel ensemble learning tool for precise prediction of proinflammatory peptides (PIPs), constructed by fusing 50 independent Random Forest (RF) models. Addressing dataset imbalance, random under-sampling generates 10 balanced model groups trained on 5 feature subsets (amino acid composition, dipeptide composition, composition-transition-distribution, physicochemical properties, amino acid index), each with 10 RF models.