Antimicrobial Peptide Assay · In Silico

Minimum Inhibitory Concentration Predictor

Submit a peptide sequence and a target organism. The model estimates the MIC needed to inhibit visible growth of that strain.

Bachelor's degree project Trained on DBAASP data Research use only

Specimen

Assay readout

Results will appear here once you run a prediction.

How it works

What to enter, and what you get back

1

Enter a peptide sequence

Type or paste an amino-acid sequence into the Peptide sequence box using the standard one-letter codes (ACDEFGHIKLMNPQRSTVWY) — e.g. GIGKFLHSAKKFGKAFVGEIMNS. Lowercase is fine, it’s uppercased automatically. Sequences work best between roughly 5–60 residues; much shorter or longer ones still run but come back with a reliability warning.

2

Pick a target organism

Choose the bacterial or fungal strain you want the MIC estimated against from the Target organism dropdown, then hit Run prediction. The model looks at that organism’s taxonomy and Gram status alongside the sequence — the same peptide can get very different MIC estimates against different organisms.

3

Read the assay readout

The big number is the predicted MIC in µg/mL — the lowest peptide concentration expected to stop visible growth of that organism. Lower means more potent. Underneath it, log10(MIC) is the raw value the model was trained to predict; the µg/mL figure is just 10 raised to that number, shown because it’s the unit used in real susceptibility assays.

4

Check the residue chain & warnings

The colored chain below the MIC value shows each residue’s chemical class (basic, acidic, polar, hydrophobic, aromatic, structural) so you can eyeball the sequence’s composition. Any amber warnings mean the sequence falls outside the typical range of the training data — treat that prediction with more caution.

A real, held-out example the model was not trained on

RWLRLNGRWLRL vs. E. coli (ATCC 25922)

Sequence RWLRLNGRWLRL
1.15 log10(MIC) = 14.1µg/mL

RWLRLNGRWLRL is a sequence from DBAASP’s held-out test set — it never appears in the model’s training data. Seven independent lab assays measured its MIC against E. coli ATCC 25922 at between 7.7 and 31.1 µg/mL; the median of those measurements is ≈14.3 µg/mL (log10(MIC) ≈ 1.15). Run this same sequence and organism through the model above and it predicts ≈14.1 µg/mL — within a couple percent of the real assay result. In plain terms: a concentration of about 14 micrograms of peptide per milliliter of broth was the lowest dose that stopped visible E. coli growth in these experiments, and that’s what the model estimated too. A lower number means the same thing but with less peptide needed — e.g. a predicted log10(MIC) = 0.56 corresponds to MIC ≈ 100.56 ≈ 3.6 µg/mL, a more potent result than this peptide against the same strain; log10(MIC) = 2.0 (MIC = 100 µg/mL) would be a weaker one. Predictions won’t always land this close — see the project journey page for the model’s overall accuracy across the full test set.