AI Peptide Design, Explained
TL;DR: AI peptide design turns a plain-English research goal into novel candidate peptide sequences in under a minute, scoring each for properties such as binding and stability. It is a discovery tool for researchers, not a source of therapeutics. Everything Velox supplies is a research reagent for in vitro lab use only.
AI peptide design turns a plain-English research goal into novel candidate sequences in under a minute. Here is how it works, what the scores mean, and where it fits in a research workflow.
What is AI peptide design?
AI peptide design uses a language model trained on peptide chemistry to propose new amino-acid sequences that fit a described research goal. Instead of searching a catalogue, you describe what you want to study in plain English and the model generates candidate sequences built for that target.
Velox Design Lab is the first tool of its kind aimed at research-peptide workflows: describe a target, get three novel candidates, each scored and novelty-checked, in under a minute.
From plain English to a scored candidate
You do not need notation or a chemistry background. You type something like "a GLP-1, GIP and glucagon triple agonist optimised for synthesisability", and the tool interprets that into a structured brief, then proposes sequences.
Each candidate is scored on synthesisability, novelty and length-optimisation, and labelled as likely-novel, partially similar, or already-existing. You can read more on what makes a peptide novel.
Where AI design fits in a research workflow
AI design is an idea-generation step. The outputs are computational hypotheses — proposals for in-vitro evaluation, not validated compounds. They give a researcher a structured starting point and a spread of distinct candidates rather than a single guess.
When a research target overlaps a well-characterised compound, it is usually faster to study the known material directly. For metabolic and incretin research, for example, that often means retatrutide; for tissue-repair work, recovery compounds such as BPC-157 and TB-500.
How to read the three scores
Every candidate comes back with three numbers, and each one answers a different practical question before you decide whether a sequence is worth synthesising.
Synthesisability estimates how straightforward the sequence would be to make on standard solid-phase equipment. Certain residues, long hydrophobic runs and awkward folding patterns lower the score because they raise the risk of failed couplings or poor yield. A high score means the chain is more likely to come off the synthesiser clean and in usable quantity.
Novelty places the sequence against a reference set of catalogued peptides — the deeper explainer is in what makes a peptide novel. A high novelty score means the exact chain has not been described before; a low score means you are effectively looking at a known compound under a new name.
Length-optimisation rewards sequences that are no longer than they need to be. Shorter, well-formed chains are cheaper to make, easier to purify and more stable in solution, so a design that achieves its intended geometry with fewer residues scores higher than a bloated one.
The three often pull against each other — the most novel candidate is not always the most synthesisable — so the value of seeing all three at once is that you can pick the trade-off that suits your budget and your research question rather than chasing a single metric.
A worked example
Say you type: "a GLP-1, GIP and glucagon triple agonist optimised for synthesisability". The tool parses that into a structured brief — three target receptor families, a stated preference for ease of manufacture — and returns three distinct candidate sequences rather than one.
Candidate A might be flagged partially similar to an existing incretin agonist with high synthesisability; a sensible prompt to study the well-characterised material directly, which for incretin work often means retatrutide. Candidate B might score genuinely novel but carry a lower synthesisability score, marking it as a more ambitious, higher-risk make. Candidate C might sit in between. None of the three is a recommendation or a validated compound — they are computational starting points for in vitro evaluation, and the labels simply help you decide which hypothesis is worth the bench time.
What it is not
AI peptide design does not produce medicines and makes no health, medical or dosing claims. Generated sequences have not been synthesised, tested, or evaluated by any regulator. Everything is supplied strictly for in vitro research by qualified researchers.
