The Antimicrobial Peptide Database (APD): A Research Guide
Peptide Databases
3,379 AMPs
The Antimicrobial Peptide Database catalogs 3,379 natural antimicrobial peptides from bacteria, plants, and animals, with structural data, activity annotations, and tools for peptide design.
Wang et al., Protein Science, 2023
Wang et al., Protein Science, 2023
View as imageThe Antimicrobial Peptide Database (APD) has been online since 2003, making it one of the oldest continuously maintained peptide resources in the world. Built by Guangshun Wang at the University of Nebraska Medical Center, the APD focuses exclusively on natural antimicrobial peptides with defined sequences and demonstrated antimicrobial activity. It is not a collection of everything that might kill a bacterium. It is a curated catalog of peptides that nature already tested over hundreds of millions of years of evolution.[1] For the broader landscape of peptide databases, see our overview of where to find every known bioactive peptide.
Key Takeaways
- The APD catalogs 3,379 natural antimicrobial peptides as of its latest update, with 6,309 total entries including synthetic and predicted peptides (Wang et al., 2023)
- Entries span six kingdoms of life: 261 bacteriocins from bacteria, 4 from archaea, 7 from protists, 13 from fungi, 321 from plants, and 1,972 animal host defense peptides (Wang et al., 2016)
- The database includes 2,169 antibacterial, 959 antifungal, 172 antiviral, 105 anti-HIV, 80 antiparasitic, and 185 anticancer peptides
- 351 unique three-dimensional structures are cataloged, 307 determined by NMR spectroscopy and 44 by X-ray crystallography (Wang et al., 2016)
- Machine learning models trained on APD data have discovered novel AMPs from the global microbiome, identifying 863,498 candidate peptides from which validated hits were derived (Santos-Junior et al., 2024)
- A database-guided, non-machine-learning approach using APD statistics designed potent, non-hemolytic AMPs in a single design step (Mechesso et al., 2026)
What the APD contains and how it is organized
The APD focuses on natural antimicrobial peptides, meaning peptides isolated from living organisms that have demonstrated activity against at least one microbial target in laboratory assays. This scope is deliberately narrower than some competing databases. The APD does not include all synthetic peptides or computationally predicted peptides in its core dataset (though synthetic and predicted entries have been added in later versions).[2]
Each entry includes:
- Amino acid sequence and length
- Source organism and kingdom of origin
- Activity spectrum: antibacterial, antifungal, antiviral, antiparasitic, anticancer, and newer annotations including antibiofilm, wound healing, and antioxidant activities
- Structural class: alpha-helical, beta-sheet, mixed, or extended/coil
- Three-dimensional structure when available (351 structures in the database)
- Minimum inhibitory concentration (MIC) data where published
- Post-translational modifications including disulfide bonds, amidation, and lipidation
The taxonomic distribution reveals where nature concentrates its antimicrobial defenses. Of the natural AMPs in the APD, 1,972 (the majority) come from animals, particularly from amphibians, insects, fish, and mammals. Plants contribute 321 entries. Bacteria contribute 261 bacteriocins. Fungi, archaea, and protists together account for only 24 entries, not because they lack antimicrobial peptides but because they have been less systematically studied.
The APD's evolution: from APD1 to APD6
The database has gone through major version expansions:[1]
APD1 (2003): Launched with several hundred entries. The original database focused on establishing a standardized format for antimicrobial peptide data.
APD2 (2009): Expanded entries and added classification tools, prediction calculators, and search functionality.
APD3 (2016): Reached 2,619 AMPs with comprehensive activity annotations.[2] Added antibiofilm, antimalarial, insecticidal, spermicidal, chemotactic, wound healing, antioxidant, and protease inhibitor annotations. The APD3 paper in Nucleic Acids Research became the most-cited version.
APD6 (2026): Contains 6,309 peptides total, including 3,379 natural AMPs, 2,290 synthetic AMPs, and 373 predicted AMPs. The expansion to include synthetic and predicted peptides reflects the growing role of computational design in the field.
The twenty-year anniversary review (Wang, 2023) documented how the APD has been used in over 2,000 publications, making it one of the most influential resources in antimicrobial peptide research.
How researchers use the APD
Drug discovery and peptide design
The APD's primary value is as a starting point for designing new antimicrobial agents. Researchers query the database for peptides active against their target pathogen, identify structural features associated with potency, and use those features as templates for synthetic optimization.
Mechesso et al. (2026) demonstrated this approach directly: using APD statistical parameters (amino acid frequencies, charge distributions, hydrophobicity profiles) without any machine learning, they designed potent, non-hemolytic antimicrobial peptides in a single design step.[3] The APD provided the design rules; the researchers applied them. This "database-guided" approach is faster than screening random peptide libraries and more interpretable than black-box machine learning.
Hilpert et al. (2008) pioneered systematic approaches to AMP screening and optimization using database-derived principles, demonstrating that short linear cationic AMPs could be rapidly optimized using structure-activity relationships derived from peptide databases.[4]
Machine learning training data
The APD has become one of the primary training datasets for machine learning models that predict antimicrobial activity. The database provides curated positive examples (peptides with confirmed activity) that models use to learn what makes a peptide antimicrobial.
Santos-Junior et al. (2024) published in Cell a machine learning approach that mined the global microbiome for novel antimicrobial peptides. Their model, trained in part on APD data, identified 863,498 candidate AMPs from metagenomic sequences. From these candidates, they validated novel peptides with activity against clinically relevant pathogens including drug-resistant strains.[5]
Wan et al. (2024) reviewed the landscape of machine learning for AMP identification and design in Nature Reviews Bioengineering, noting that database quality is the limiting factor for model performance. The APD's curation standards make it particularly valuable as training data because the entries have verified sequences, confirmed activities, and standardized annotations.[6]
Lee et al. (2023) applied BERT (a transformer-based language model) to predict AMP function, demonstrating that natural language processing architectures can learn AMP properties from database-derived sequences.[7]
For readers interested in how machine learning is revolutionizing antimicrobial peptide prediction, the APD is often the foundation.
Structure-activity analysis
With 351 three-dimensional structures, the APD enables systematic analysis of how peptide structure relates to antimicrobial function. Most natural AMPs adopt amphipathic structures: one face is hydrophobic (interacts with the lipid membrane of bacteria) and the opposite face is cationic (attracted to the negatively charged bacterial membrane surface). The APD quantifies these properties for each entry, enabling researchers to identify the structural features that distinguish highly active AMPs from weak ones.
Henson et al. (2025) used database-derived design principles to create novel tryptophan- and arginine-rich AMPs with high solubility and low red blood cell toxicity, directly addressing two of the main barriers to clinical AMP development.[8]
APD vs. other antimicrobial peptide databases
The APD is not the only AMP database. Two major alternatives serve different needs:
DRAMP (Data Repository of Antimicrobial Peptides): Contains over 30,000 entries as of version 4.0, including general AMPs, patent AMPs, clinical AMPs, and stapled AMPs. DRAMP is larger than the APD because it includes patent literature and clinical pipeline entries. Its focus on clinical translation makes it valuable for pharmaceutical researchers.
DBAASP (Database of Antimicrobial Activity and Structure of Peptides): Contains over 22,000 entries with detailed activity data against specific bacterial strains. DBAASP's strength is quantitative activity reporting (MIC values against defined pathogens), making it useful for structure-activity modeling.
The APD's comparative advantage is curation quality and the focus on natural peptides. Every natural AMP in the APD has a verified sequence and published activity data. The database is smaller than DRAMP or DBAASP but more stringently curated. For researchers who need clean, reliable training data for machine learning models, the APD's curation standards matter more than raw entry count.
Marciano et al. (2025) published a comprehensive overview of antimicrobial peptides including computational approaches, noting that the choice of database affects the quality of downstream analyses.[9]
Why natural AMPs matter in the antibiotic resistance era
The APD exists because of a clinical crisis. Antibiotic-resistant bacteria are an accelerating global health threat. The WHO lists antimicrobial resistance among the top ten global public health threats. Traditional small-molecule antibiotics are failing against MRSA, carbapenem-resistant Enterobacteriaceae, and extensively drug-resistant tuberculosis.
Natural antimicrobial peptides represent an alternative approach. They kill bacteria through membrane disruption, a mechanism fundamentally different from conventional antibiotics. Because AMPs attack the physical structure of bacterial membranes rather than specific enzymatic targets, resistance development is slower and harder. Bacteria can develop resistance to AMPs, but it requires wholesale membrane restructuring rather than a single point mutation.
Babuccu et al. (2025) used machine learning trained on AMP database data to identify peptides active against multidrug-resistant bacteria, demonstrating the direct clinical relevance of database-derived peptide discovery.[10]
The APD catalogs the raw material for this approach: thousands of peptides that evolution has already optimized to kill bacteria. The challenge is not finding active peptides (the APD has thousands). The challenge is making them stable enough, non-toxic enough, and affordable enough for clinical use. That translation gap, from database entry to approved drug, remains the field's central problem.
For readers interested in how immunoinformatics predicts peptide immune responses or how computational toxicity testing works, these tools often draw on the same database infrastructure as AMP discovery.
The Bottom Line
The Antimicrobial Peptide Database (APD) has cataloged over 3,379 natural antimicrobial peptides across six kingdoms of life since 2003, with the latest version expanding to 6,309 total entries. It serves as both a research reference and a training dataset for machine learning models that discover novel antibiotics. The APD's curated entries have supported peptide design, structure-activity analysis, and the identification of hundreds of thousands of candidate AMPs from the global microbiome. While smaller than competing databases like DRAMP, the APD's curation quality makes it a preferred resource for computational drug discovery in the antibiotic resistance era.