Markush Claims and AI: What Changes, and What Doesn't

• Generating a Markush formula is the easy part; knowing which compounds it still covers after each amendment is the work.
• At the EPO, even narrowing R1 to hydrogen is a selection: build fallback sub-formulae at drafting, not during prosecution.
• Software can run the coverage check; which compounds matter, and how much breadth to defend, stays with the attorney and the client.

A Markush claim is rarely lost at filing. It is lost a little at a time, during prosecution: a substituent list shortened to answer a novelty objection, a proviso added to get past a reference, a species elected while the rest of the genus waits. Each move is reasonable on its own. What rarely gets asked after each one is the question put to attendees of our September session with The Patent Lawyer: how do you check that the compounds you intended to protect are still covered by the claim you end up with?

‍

In most practices the honest answer is: by reading. The attorney who knows the file holds the claimed chemical space in mind and re-derives it after each amendment. That works for a careful practitioner on a familiar case. It works less well on a file inherited from a colleague, on the fourth amendment, or when the compound the client cares about most is not the one that was elected.

‍

The early poll answers in the session pointed the same way: coverage is mostly checked by hand. Peter Mumford of Potter Clarkson, who drafts new chemical entity cases for early-stage biopharma clients, described his own routine. Every time he advises on a narrowing amendment, he asks which compounds the client would be giving up, because the answer can drive divisional strategy. With two hundred compounds on file, pragmatism wins: few clients want to pay for all two hundred to be checked by hand, but they do want their ten most important compounds checked every time.

‍

This piece is about that gap: where coverage is lost, what each office's rules make you watch for, and which part of the check belongs to software and which part does not.

‍

A Markush claim is two objects that only mean something together

‍

A generic formula and the definitions of its variables describe one thing, but they live in different places: the formula as a drawing, the definitions as text, sometimes pages apart, with provisos and dependencies between substituents. Coverage is a property of the combination. Narrow R² from C₁-C₆ alkyl to C₁-C₃ alkyl and the drawing does not change, yet every compound carrying a butyl, pentyl or hexyl at that position has just left the claim.

The practical consequence is simple and easy to underestimate: coverage has to be re-established after every change to either side, not assumed from the previous version.

‍

Where compounds fall out of scope

• Shortened substituent lists. The most common response to a novelty objection over a reference that discloses part of the genus.
• Provisos and disclaimers. Each one carves out a region of the chemical space, sometimes wider than intended once it interacts with other definitions.
• Election of species in the US. Under MPEP 803.02, the examiner may require an election and search the elected species first, extending the search only if it is allowable. Compounds covered only by the non-elected part of the genus depend on decisions made later, often in a continuation.
• Improper Markush grouping. Under MPEP 2117, members of a Markush group must share a single structural similarity and a common use. Responses to this rejection usually narrow the group.
• Sub-formulae built as fallback positions. Useful in prosecution, but each layer has its own coverage, and the compounds that matter are not always in the layer that survives.
• A lead that moves. The compound that ends up mattering commercially can change after filing. Mumford sees clients whose lead is not exemplified at all but came out of later lead optimization, which makes generic coverage, not just the examples, worth tracking through prosecution.

In each case the compounds to track are the same: the lead candidates, the compounds in the examples, and the ones the client may develop next.

‍

The same amendment does not carry the same risk at each office

‍

United States

‍

A genus claim needs written description support under 35 U.S.C. §112(a): a representative number of species or structural features common to the members of the genus (Ariad Pharmaceuticals v. Eli Lilly, 598 F.3d 1336 (Fed. Cir. 2010) (en banc)). It also has to be enabled across its full scope: Amgen Inc. v. Sanofi, 598 U.S. 594 (2023), confirmed that a claim to an entire class must enable the entire class. Provisos are negative limitations and need a basis in the original disclosure (MPEP 2173.05(i)). And every narrowing amendment becomes part of the prosecution history, with the presumption of surrender of equivalents set out in Festo, 535 U.S. 722 (2002).

‍

European Patent Office

‍

Amendments must be directly and unambiguously derivable from the application as filed (Art. 123(2) EPC). In Markush practice, the recurring trap is selection from two lists: picking specific items out of two separate lists at once can create a combination that was never disclosed as such, while deleting items from a single list is generally treated differently.

‍

As Mumford put it in the session, even narrowing R¹ from a list to hydrogen is already a selection, which is why he builds sub-formulae (1A, 1B, 1C) into the application at drafting, so the fallbacks exist before they are needed. Undisclosed disclaimers are allowed only within the limits of G 1/03, for instance to restore novelty over Art. 54(3) prior art or an accidental anticipation, and must not contribute to inventive step. Disclosed disclaimers fall under the standard confirmed in G 2/10. Each proviso added also has to leave the claim clear under Art. 84.

‍

The consequence for a family filed in both offices: an amendment that is unremarkable in the US can be added matter in Europe, and the reverse. Coverage has to be checked per jurisdiction, against the claim as it stands in that jurisdiction.

‍

Five checks after every amendment

‍

1. Write down the compounds that must stay covered before drafting the amendment, not after: leads, examples, and anything the client has flagged.
2. Check each one against the amended claim, not the version you remember.
3. Decide where each compound that falls out goes: a dependent claim, a divisional or continuation, or consciously given up. Make it a decision, not a side effect.
4. Check the basis office by office: written description for provisos in the US, the two-list question and disclaimer limits at the EPO.
5. Record why. The file history will be read by someone else one day, possibly in litigation.

‍

What software can check, and what stays with the attorney

‍

Whether a specific structure falls within a genus as defined has a correct answer. It should be computed, not estimated, and it should give the same answer every time it is asked. That is the part of the check where software helps most: showing which compounds left the claim after an edit, across every version of the claim and every jurisdiction, without relying on anyone's memory of the file.

‍

What it cannot decide is which compound matters commercially, how much breadth is worth defending, or what risk the client will accept. That information is not in the chemistry. It sits with the attorney and the client, and so does the responsibility for the claim that goes on file.

Mumford was clear on where he draws the line. He would want to run the coverage check himself at least once, because as the attorney legally responsible for the application, he has to be able to stand behind a statement that every compound is covered.

‍

This article covers drafting practice and tooling. It is not legal advice.

Quentin Jorget, European Patent Attorney, Patent Solutions Engineer, DeepIP

Peter Mumford, Senior Associate, Patent Attorney, Potter Clarkson