🔬 AMP Candidates

2,065 computationally predicted antimicrobial peptide candidates

🧬 About these candidates

Each candidate was identified by mining extreme-environment metagenomes with ESM-2, a protein language model trained on 250M sequences. Candidates pass multi-stage filtering: biophysical scoring, novelty screening against 863K known AMPs (APD3 + DRAMP + AMPSphere), hemolysis risk prediction, and structural validation via AlphaFold2.

⚠️ All candidates are computationally predicted — no experimental validation has been performed.

🔒Amino acid sequences are not shown — sequences are proprietary research data. Interested in collaboration? Contact us.
🌋
#2065

🌋 Hot Springs

database-novel
ESM-2 AMP ScoreProbability of antimicrobial activity (0–1). Higher = more likely AMP.

0.8702

LengthPeptide length in amino acids40 aa
Net ChargeCharge at pH 7. Most AMPs are cationic (+)2.5
HydrophobicFraction of hydrophobic residues (0–1)45%
PhyschemComposite physicochemical plausibility score5.0
✅ Low hemolysisPredicted probability of red blood cell lysis. <0.5 = non-hemolytic(0.48)
0% max identityHighest sequence identity to any known AMP in databases
← PreviousShowing 20652065 of 2,065

📚 Understanding the metrics

ESM-2 AMP Score: Probability (0–1) that the peptide has antimicrobial activity, predicted by Meta's ESM-2 protein language model.
Net Charge: Charge at physiological pH. Most effective AMPs carry a positive charge (+2 to +9).
Hemolysis Prob: SVM-predicted probability of red blood cell lysis (HemoPi3). Below 0.5 is considered non-hemolytic.
Novelty Tier: How similar to known AMPs. "database-novel" means <50% identity to any AMP in APD3 + DRAMP + AMPSphere.
Hydrophobic fraction: Proportion of hydrophobic amino acids. AMPs typically have 40–60% for membrane interaction.
Cross-biome: Found independently in multiple extreme environments — suggests evolutionary conservation of antimicrobial function.