Peptide versus strain screening

Antimicrobial peptide prediction tool: predict AMP activity and MIC per strain

Paste a sequence, pick a strain panel, and get a predicted MIC band per organism with the expected mechanism named. Classifiers tell you a peptide is probably antimicrobial. This tells you what it is likely to be antimicrobial against.

Plans from $149 a month

Sequences stay yours
Never used for training
No card required

Screen a peptide

of 3 left

Peptide, by sequence or name

Strain panel

S. aureus (MRSA) E. coli K. pneumoniae (CRE) A. baumannii P. aeruginosa E. faecium (VRE)

Read-out

Include closest published analogs

No account needed. 3 screens per session.

What comes back

Rows: One per strain in the panel
Per row: MIC band, S I R call, mechanism
Support: Confidence 1 to 4 and analogs
Typical run: About ten seconds

Ready to run Colistin ESKAPE panel

Nothing predicted until you run

Strain	MIC (µg/mL)	Call	Resistance risk	Conf.
	—	Not run	Predicted when you run the screen.
Tick at least one strain in the panel on the left.

Nothing here is predicted yet

These are the strains this page preselected, waiting on a compound. Run the screen and every row fills in with an MIC band, an S / I / R call, the mechanism expected to decide it and a confidence out of four.

Strain	MIC (µg/mL)	Call	Resistance risk	Conf.

Why

Closest published analogs

Want the rest of your series? Batch runs are on the paid tiers. screen left in this session. This session is used up.

Computational prediction from published literature. Research use only, not a lab measurement and not clinical guidance.

The short answer

What an antimicrobial peptide prediction tool should return

An antimicrobial peptide prediction tool estimates how a peptide will behave against bacteria from its sequence, before any synthesis or plate work. Most of the published tools are classifiers: they return one probability that the sequence is antimicrobial at all. That is a useful filter when you are mining a proteome for candidates, and it is the wrong output once you have a series and a decision to make.

Antibacterial returns the other answer. You paste a sequence, a peptide name or a structure, pick a strain panel, and get a peptide by strain matrix: a predicted MIC band in µg/mL per organism, an S, I or R call, the mechanism the model expects to limit it, the closest characterised peptides the call rests on, and how much published support each row actually has. A novel sequence with no measured relatives comes back marked no basis rather than dressed up as a number.

At a glance

Input: A peptide sequence, a peptide name, or a SMILES string
Output: Predicted MIC band per strain, S/I/R call, mechanism
Not: A yes or no "is this an AMP" classifier score
Grounding: Published activity for the closest characterised peptides

Two different outputs

A classifier score and a predicted MIC are not the same product

This trips up more peptide programs than any other single thing. Both outputs are called prediction, both come from sequence, and they answer questions that sit at opposite ends of the project.

	AMP classifier	Per-strain activity prediction
Question answered	Is this sequence antimicrobial?	What is it likely to inhibit, and at what concentration?
Output	One probability, typically 0 to 1	A MIC band per organism, with an S, I or R call
Organism named	No	Yes, per strain in the panel
Typical input scale	Thousands of sequences from a proteome	A designed series of ten to a few hundred
Where it fits	Discovery, before you have candidates	Triage, once you have candidates and finite plate capacity
Examples	amPEPpy, CAMP, iAMPpred, APD tools, PyAMPA	Antibacterial
Cost	Free, mostly academic and open source	From $149 a month, published

The two are complements, not rivals, and the free classifiers are genuinely good at their job. Run a classifier when you are mining a genome for candidate peptides. Run a per-strain prediction when you have forty designed analogs, plate capacity for eight, and a project meeting on Thursday.

What you get back

Six columns, and one of them is the confidence

01

Predicted MIC band

A range in µg/mL per organism rather than a single value. Broth microdilution reads in doubling dilutions, so the bench answer is plus or minus one step anyway and a band is the honest unit.

02

S, I or R call

Read against the published breakpoint for that species where one exists. Peptides frequently have no breakpoint at all, and the row says so instead of inventing a category.

03

Expected mechanism

What the model expects to limit the peptide: outer membrane exclusion, lipopolysaccharide binding, protease cleavage, or salt sensitivity at physiological cation concentration.

04

Closest characterised peptides

The relatives the call rests on, with their measured activity. You can check the reasoning against the literature instead of trusting a score you cannot audit.

05

Confidence level

How densely measured that neighbourhood is. A novel motif gets low confidence and a wide band. That is the correct output, not a failure of the tool.

06

Export and API

CSV on every plan, structured reports on Discovery and above, and API access for teams that want the read-out inside their own design loop.

How a screen runs

Four steps, and the whole thing takes seconds

STEP 1

Paste the sequence

Single letter amino acid code, a peptide name such as LL-37 or nisin, or a SMILES string for a cyclised or non-canonical peptide.

STEP 2

Pick the panel

ESKAPE, Gram-negative, Gram-positive, or a custom list of the isolates your program actually cares about.

STEP 3

Read the matrix

One row per strain: band, call, mechanism, analogs, confidence. The rows with thin support are labelled as thin, not quietly averaged.

STEP 4

Rank and cut

Export the matrix, order the series, and send the top few to synthesis or assay with a written reason for every peptide you left out.

Who it is for

Four people who ask this question for a living

Peptide chemist on an AMP program

You have forty analogs off a magainin or cathelicidin template and capacity to synthesise eight. A per-strain ranking with the mechanism named tells you which eight, and gives you a defensible reason for the other thirty-two.

Head of discovery at a peptide biotech

You own the go or no-go on a template. Killing a series that will never cross the Gram-negative outer membrane at week two rather than month five is the largest line item you control.

Computational biologist supporting a wet lab

You already run classifiers over the proteome and the chemists keep asking the follow-up question you cannot answer from a probability score. This is the layer that answers it, with an API so it sits inside your existing loop.

Academic AMR lab principal investigator

Synthesis and assay slots are the scarce resource on a fixed grant. A literature-grounded ranking is the evidence that decides which peptides get the limited bench time this quarter.

Strain panels

Peptides fail per organism, so the read-out is per organism

A sequence-level score averages away the thing that decides a peptide program. The same peptide can be potent against MRSA and thirty-two fold weaker against Gram-negative bacteria, for reasons that live in the outer membrane rather than in the peptide. The background on the mechanism and liabilities of antimicrobial peptides is written up separately.

ESKAPE

Six pathogens behind most resistant hospital infections.

S. aureus (MRSA)
E. coli
K. pneumoniae (CRE)
A. baumannii
P. aeruginosa
E. faecium (VRE)

Gram-negative

Permeability and efflux decide almost everything here.

E. coli
K. pneumoniae (CRE)
A. baumannii
P. aeruginosa
E. cloacae
S. Typhimurium

Gram-positive

Target modification and acquired operons dominate.

S. aureus (MRSA)
S. aureus (MSSA)
E. faecium (VRE)
E. faecalis
S. pneumoniae
S. epidermidis

Custom panels and your own clinical isolates are available on the Discovery plan and above.

What it will not do

The limits, before you pay for it

A prediction is for deciding what to make and what to test. It is not a measurement, it never becomes the number in a paper or a filing, and it is worth knowing exactly where it stops before a plan depends on it.

It does not replace broth microdilution

The measured MIC against your isolates is the ground truth and stays the ground truth. Prediction decides which peptides are worth that plate, nothing more.

Novel motifs return low confidence, on purpose

If a sequence has no characterised relatives, there is no honest band to give. The row says no basis. That is information, and it is the row worth a plate.

In vivo behaviour is a separate problem

Serum stability, renal clearance and toxicity are not predicted from a MIC and are not claimed here. A potent peptide with a two minute half life is still a dead peptide.

Non-canonical chemistry narrows the record

D-amino acids, cyclisation, lipidation and unnatural residues all shift activity in ways the published set covers unevenly. Confidence drops accordingly, and the read-out says so.

Questions people ask before buying

Antimicrobial peptide prediction, answered plainly

What is the best antimicrobial peptide prediction tool?

It depends on which question you are asking. If you need to know whether a sequence is antimicrobial at all, an open classifier such as amPEPpy, CAMP or the APD tools answers that well and costs nothing. If you need to know which organisms a peptide is likely to inhibit and at roughly what concentration, you need a per-strain activity read-out instead, which is what this tool returns.

Can antimicrobial peptide activity be predicted from sequence?

Activity can be predicted as a band, not as an exact number. Sequence carries most of the signal that matters: net charge, hydrophobic moment, helicity and length drive membrane selectivity, and those features have been mapped against measured MICs for thousands of peptides. A prediction is reliable in proportion to how many characterised relatives your sequence has.

How accurate is antimicrobial peptide prediction?

Accuracy tracks the density of the published neighbourhood, not the sophistication of the model. A magainin or cathelicidin analog sits in a crowded, well measured region and gets a tight band with high confidence. A genuinely novel scaffold gets a wide band and low confidence, which is the correct answer. Any tool returning the same confident number for both is hiding what you need to know.

What is the difference between an AMP classifier and MIC prediction?

A classifier returns a probability that a sequence belongs to the antimicrobial class, usually a number between 0 and 1. It does not name an organism and it does not name a concentration. MIC prediction returns a band in µg/mL against a specific strain, with an S, I or R call. Classifiers filter a proteome. MIC prediction decides which peptides earn plate time.

Can I screen a peptide against Gram-negative bacteria?

Yes, and it is the case worth screening first. Most peptides that look excellent on a Gram-positive plate lose an order of magnitude against Pseudomonas or Acinetobacter, because the outer membrane and lipopolysaccharide layer change the binding problem entirely. Getting that gap flagged before synthesis is where the plate time is actually saved.

Does the tool flag haemolysis and protease stability?

It flags what the literature reports for close relatives, which is the honest limit. Haemolytic activity and serum protease susceptibility are the two liabilities that kill most peptide programs, and both are strongly conserved within a structural family. The read-out names them where characterised analogs show them, and says nothing where the record is silent.

Do my peptide sequences stay confidential?

Sequences you submit stay yours and are never used to train models. That matters for an unpublished series, and it is the practical reason a company cannot simply route everything through a free academic screening scheme, where results enter an open database after an embargo period.

Pricing

What peptide activity prediction costs

Full plan comparison

Published, because a quote cycle costs a small team weeks it does not have. If you also screen small molecules, the same account covers both: the MIC prediction software read-out is the same matrix with a structure as the input, and the antibacterial screening API returns it as JSON for a design loop that never opens a browser.

Yearly billing is two months cheaper, charged once a year.

Lab

$124 /mo

$1,488 charged once a year

250 screens a month against the standard panels, with CSV export.

Get started

Academic AMR lab, 1 to 2 chemists

Discovery

Screen a peptide before you commit the synthesis slot

Three screens per session, no account needed. Paste a sequence, pick a panel, read the matrix.

Run a screen first

What you are agreeing to

Your compounds stay yours
Never used for model training
Deletable on request
No card needed to create your account