AntiBP2: improved version of antibacterial peptide prediction

Welcome to the official repository for AntiBP2, an improved and robust computational method for predicting and classifying antibacterial peptides (ABPs) from amino acid sequences. This resource is designed to support researchers in antimicrobial peptide therapeutics, innate immunity, and computational drug discovery.

Web Server: https://webs.iiitd.edu.in/raghava/antibp2/

Citation Lata, S., Mishra, N. K., & Raghava, G. P. S. (2010). AntiBP2: improved version of antibacterial peptide prediction. BMC Bioinformatics, 11(Suppl 1):S19. https://doi.org/10.1186/1471-2105-11-S1-S19

About the Tool AntiBP2 is an updated version of the original AntiBP method, developed to predict and classify antibacterial peptides using Support Vector Machine (SVM) models trained on a significantly larger dataset. It consolidates sequence-level features into a unified prediction framework, enabling systematic identification of antibacterial peptides from raw amino acid sequences.

The tool integrates data from:

Antimicrobial Peptide Database (APD) Swiss-Prot / UniProt MitPred dataset (for negative peptide generation) Key Features Large Dataset

999 unique antibacterial peptides (positive set) 999 randomly extracted non-antibacterial peptides (negative set) Nearly double the dataset size compared to the original AntiBP Multiple Prediction Models

NT15 — binary pattern of first 15 N-terminal residues CT15 — binary pattern of last 15 C-terminal residues NTCT15 — combined N- and C-terminal binary patterns Whole Peptide — amino acid composition of full-length peptide Source Classification

5 biological sources: Bacteria, Frog, Insects, Mammals, Plants Overall classification accuracy up to 98.95% Subfamily Classification

Insect families: Apidaecin, Attacin, Cecropin, Invertebrate Defensin, Lebocin Mammalian families: Alpha-defensin, Beta-defensin, Cathelicidin, Hepcidin, Histatin Frog families: Bombinin, Brevinin, Caerin, Dermaseptin, Other Rich Performance Metrics

Sensitivity, Specificity, Accuracy, and MCC for every model Validated on an independent blind dataset of 466 sequences