How do I develop a nanobody against an epitope already claimed by a competitor's patent?

Written by the Immunochem scientific team and reviewed by Philippe Herman, CEO and Chief Scientific Officer. Published 13 October 2026, updated 13 October 2026.

If you are still at the earlier question, whether the epitope is reachable at all and what the two routes are, start with designing around an antibody patent. This page assumes you have decided to go ahead.

What we need from you before anything starts

Four things, and you probably have three of them already.

The patent number, or the sequences themselves. Published claims list the CDRs or the full variable domains, and those sequences are the reference we measure against. A structure of the target, experimental or predicted. An AlphaFold model is usually workable, and if the epitope is defined by a co-crystal with the competitor's antibody, the PDB entry is better still. An indication of which epitope you need to engage, in whatever form you have it, whether that is a residue list, a published figure or a functional description. Finally, the threshold your attorney wants us to stay below, if they have already set one. If they have not, we start without it and the number can be applied later.

What we do not need is an animal. We never run immunisation campaigns, of any species. The sequences come from computational design against the structure, which is also why the prior art question stays clean: there is no library, no immune repertoire and no selection step whose provenance anyone has to trace.

Step one: the free feasibility study

Before you commit to anything, we run a feasibility study on your specific site and return a compatibility score for it. It tells you whether that epitope is one we can realistically generate credible binders against, because not every surface is. Flat, featureless and highly polar sites are harder, and a site that has already been engaged by a published antibody is usually a good sign rather than a bad one.

The study costs nothing and carries no obligation. If it comes back negative we say so and the engagement stops there. A negative answer in two weeks is worth more than a campaign that fails in six months.

Step two: generating sequences that bind the same site differently

This is the core of the work. We know the site is bindable, because somebody has already bound it. What we do not reuse is how they bound it.

The published antibody tells us which residues carry the interaction and roughly what geometry engages them. We use that as a specification rather than as a starting scaffold: the target is the same epitope, the constraint is a different sequence. From there we generate nanobody candidates computationally and vary both the CDRs and the framework, because an identity measurement over the full variable domain counts framework positions too.

The single domain is what makes this tractable. With no pairing interface to preserve, a larger share of the nanobody sequence is free to vary, so we can put real distance between the candidate and the claims without losing the interaction. The same exercise in a full IgG has fewer free positions and takes longer.

Step three: measuring the distance to the claims

The legal question becomes a number, and the number is something we can design against.

For every candidate, we compute sequence identity to each claimed sequence, reported per CDR and over the whole variable domain, because claims are written both ways. If a claim reads "at least 90% identical to SEQ ID NO:1", your attorney tells us what margin they want below that figure and we design to it. If it lists exact CDR sequences instead, the requirement is different and we report against each one.

Every candidate ships with that table. The point is not that the table settles the legal question, because it does not. The point is that when the question is raised, two years from now, the measurement already exists and was made before the molecule was chosen rather than after.

Step four: validating in the laboratory

A designed sequence is a hypothesis until something binds. We express the candidates in a cell-free system and test them by ELISA against the target, which turns the shortlist into a set of molecules with measured binding rather than predicted binding.

This step is where we differ from platforms that deliver sequences and stop. A computational pipeline will always return something that scores well. Whether it binds is a separate fact, and it is cheaper to find out at this stage than after you have committed to a development programme.

What you receive, and when

Six weeks from go-ahead to candidate sequences. Twelve weeks to binders validated in the laboratory. Those are working timelines, not best cases.

The package is the candidate sequences, their identity to the claimed prior art computed per CDR and across the variable domain, a developability assessment covering aggregation, expression and specificity risks, the binding data from the validation round, and the design record showing how each candidate was produced and what was rejected along the way. The sequences, the data and any intellectual property arising from the work are yours.

Where our work stops and your attorney's begins

We do not give freedom to operate opinions. That is legal work and it belongs to a qualified patent attorney.

The division that works in practice is this: your counsel reads the claims and sets the threshold, and we design and measure against it. They decide what the claim covers. We decide nothing about the claim and everything about the molecule. Most clients run the two in parallel, which is faster than running them in sequence and means the threshold is known before the design work starts rather than after.

Frequently asked questions

Do you need the patented antibody's structure, or is the sequence enough?

The sequence plus a structure of the target is enough to start. A co-crystal of the two is better, because it defines the epitope precisely rather than approximately, but it is not a requirement.

Can you do this without immunising an animal?

Yes, and we do not have the option of doing otherwise. All our sequences come from computational design against the target structure. For this kind of work that is an advantage: there is no library and no selection history for anyone to examine.

What if the patent claims the epitope rather than a sequence?

That case is covered in the article on designing around an antibody patent. Two things matter here. An epitope is a natural feature of a protein, and a natural molecule is not itself patentable, so what a patent claims is an antibody defined by the epitope it binds, not the epitope. And such a claim still has to be enabled: in Amgen v. Sanofi the Supreme Court held that the broader the claim, the more the patent must teach, and Amgen's functionally defined genus fell because the 26 sequences it disclosed did not enable the class it claimed. A patent cannot reserve a broad sequence design space it has never taught how to reach, and that is what leaves room to design around it. Which kind of claim you are facing, and how strong it is, is a question for your attorney.

How many candidates do you deliver?

It depends on the site and on how tight the identity constraint is, since a strict threshold removes part of the design space. We agree the number at the start of the project rather than promising a figure here.

Can my attorney set the identity threshold themselves?

Yes, and we prefer it. They give us a figure, we design below it, and the report states the margin achieved for each candidate. The values are usually documented in the patent, but specific clauses can narrow the design space further.

Do you work on conventional antibodies as well as nanobodies?

Yes. Both routes deliver either format. A nanobody is often the easier target for this kind of work, for the reason given above, but a full IgG is a normal request and the process is the same.