A new chemical compound or molecule may look novel on paper. It may not appear in any patent database with an exact structural representation, and its IUPAC name demonstrates nothing useful. However, an earlier patent may still cover the same chemical concept through a broad Markush formula, a related scaffold, a nested R-group definition, or a series of structurally similar compounds. So how do you actually find what an exact-structure search misses?… The goal of a chemical structure search is to find patents, published applications, and non-patent literature that disclose or claim a specific molecule or a related family of molecules.
The chemical structure search focuses on the structure itself rather than on keywords. This can be used to find core scaffolds, substructures, analogues, salts, stereoisomers, and the generic Markush formulas used in patents to enable thousands (sometimes millions) of compounds to be claimed by one patent. Companies make use of this thorough search when carrying out novelty assessments, freedom-to-operate (FTO) analyses, prior-art investigations, competitive intelligence, and making R&D decisions.
Running the search is simple. But deciding what to include, evaluating the results, and knowing when to act on them require search expertise. A successful chemical structure search is not simply about finding the molecule. It is about understanding the chemical space around it. This article explains why the chemical structure search matters, how we build one in stages, where it tends to go wrong, when to run it, what AI changes, and how our IP experts handle it.
Why Is Chemical Structure Search Important?
Chemical patents are written broadly. A single claim can potentially represent an enormous number of chemical permutations, while differences in database indexing and the interpretation of structural variables can lead to missed references or noisy results. This means that a search by name or keyword can miss a patent that describes a molecule that is never named. This issue can be resolved by searching on the basis of structure. Teams do so to:
- Make sure the compound is novel before you submit it.
- Look through the patents or public disclosures that currently exist for similar compounds.
- Evaluate the risk of infringement and get the molecule cleared before launch.
- Collect prior art to support or challenge the patent claims
- Benchmark structurally related substances and identify the trends in molecular innovation.
The legal value is easiest to see in disputes that hinge on how one molecule compares with a compound in earlier patents. Two Delhi High Court rulings are appropriate examples:
- Roche v. Natco (risdiplam, October 2025). Roche sought an injunction against Natco’s generic risdiplam. The courts found risdiplam obvious over Compound 809 in an earlier genus patent, because it differed only by a nitrogen replacing a -CH group, and refused the injunction. The Division Bench upheld that on October 9, 2025, and the Supreme Court dismissed a further petition on October 17, 2025, although the main suit is still pending. What carried the argument was a close structure-to-structure comparison.
- AstraZeneca v. Intas and others (dapagliflozin, 2021). Generic makers argued that dapagliflozin was already disclosed in AstraZeneca’s earlier Markush patent. They cited a prior-art example that differed from dapagliflozin only by a methoxy versus ethoxy group, and also pointed to AstraZeneca’s own admissions in US proceedings. The court denied AstraZeneca an interim injunction and imposed costs on it.
These real-world case studies demonstrate that the outcomes depend on structure-level prior art and how a genus patent relates to a specific compound, which is precisely what a structure search is meant to identify. Since courts make a distinction between a compound being ‘covered’ by a Markush claim and being ‘disclosed’ by it, the results need expert analysis as much as they need a good search.
Best Practice: Treat structure search as risk management, not a formality before filing.
What Information Do You Need to Start?
Most gaps in search results trace back to missing or vague inputs, so it is important to have the following ready before the search begins:
1. The exact structure, with stereochemistry. Presentation should be in the form of drawing or as SMILES/InChI. These formats can be easily converted into other types of identifiers, thus avoiding the errors that occur when a name is retyped. State the stereochemistry clearly too. A patent may claim a racemate, a single enantiomer, or all stereoisomers, and that difference can decide whether a claim covers your molecule.
2. Known variants. List any salts, polymorphs, prodrugs, metabolites, and tautomers. Patents often claim these separately from the parent compound, sometimes in a different family owned by someone else. A tautomer can also be drawn in more than one way. So, a database may miss a match unless each form is searched.
3. The scope of interest. Let us know whether you are concerned about the core scaffold or about specific substituents. This determines which parts of the molecule stay fixed and which can vary in the search, and how far we go into substructure and Markush-level queries. It also indicates the positions you might plausibly alter if a blocking claim arises.
4. Intended use and territories. Note the therapeutic use, the manufacturing route, and the countries where you plan to develop, manufacture, or sell. A patent can claim a compound, a use, a formulation, or a process, so one molecule may be affected by several kinds of claims. Also, rights are territorial and a patent, which is important in one country may have no effect in another.
- The purpose of the search. Novelty, FTO, invalidity, and landscape searches each need a different strategy:
- A novelty search looks for any earlier disclosure, in patents or literature, that could affect whether the compound is new.
- An FTO search concentrates on claims that are in force in your target countries.
- An invalidity search looks for prior art that predates a specific patent, so its claims can be challenged.
- A landscape search maps the field around a molecule, scaffold, or target to show who is active and where the white space is.
| Search | Core Question | Focus |
|---|---|---|
| Novelty | Is the compound new? | Any earlier disclosure in patents or literature |
| FTO | Can we launch? | Claims in force in target countries |
| Invalidity | Can this patent be challenged? | Prior art before its priority date |
| Landscape | Who is active, and where is the white space? | The field around a molecule, scaffold or target |
5. Key dates. Provide the priority or filing date, since it sets the prior-art cut-off. For a novelty search, this is the date of your own application. For an invalidity search, it is the priority date of the patent being challenged. Recently filed applications may not be published yet, so the date also matters when reading results close to the cut-off.
How Does a Layered Chemical Structure Search Work?
One query in one database is rarely enough. An effective search is built in layers, each exploring a different part of the chemical space.
Step 1: Understand the chemistry before searching
We begin with the molecule itself, examining its core scaffold, substructures, functional groups, derivatives, isomerism, variable positions, and intended application. For large or complex molecules, the structure is bifurcated into smaller searchable fragments to improve retrieval. The important question is not just ‘What is the molecule?’ but also which parts should remain fixed and which should be permitted to vary during the search.
Step 2: Build multiple structure queries
After we identify the structure of the molecule, we create different search representations, ranging from the exact structure to the key scaffold with the attached substituents, then to the core scaffold with associated substructures. The structure can also be translated into other searchable identifiers such as the IUPAC names, SMILES, InChIKey, and CAS Registry data with the help of the Cactus/NCI Resolver.
In Markush searching, a common scaffold is kept while some designated positions are generically defined to include alternative R-groups. The question shifts from “Has this exact compound been encountered previously?” to “How widely has this fundamental chemical idea been previously disclosed?”
Step 3: Combine keyword and structure search
A comprehensive investigation uses two complementary approaches.
Keyword-based searching includes patent databases like Questel Orbit, Pat Snap, Derwent Innovation, Pat Seer, Espacenet, PATENTSCOPE, Google Patents, and USPTO data. Besides this, the search includes chemical/substructure terminology, IUPAC nomenclature, assignee and inventor data, and IPC/CPC/USPC classifications, along with backward and forward citations.
Structure-based searching then queries the chemistry directly across multiple platforms, using four levels:
- Exact structure search
- Chemical structure family search
- Substructure search
- Closed substructure search
Each level covers a distinct part of chemical space and reduces the risk of missing relevant prior art family.
Step 4: Non-patent literature alignment
Scientific literature may include important disclosable information. We search sources like Scopus, ScienceDirect, Web of Science, PubMed, PubChem, CiteSeerX, and Google Scholar, expanding by keyword, author, and affiliation. If a relevant inventor, author, university, organization, or research group is identified, these affiliations assist in finding disclosures and demonstrate how a chemical series evolved.
Step 5: Reassurance searching
A comprehensive search also tests if its own findings can be validated or extended. Reassurance searching employs two levels of forward and backward citation analysis, showing how one invention was created and what subsequent innovations are based upon it. It also includes additional searches of the assignees, inventors, and authors who were found during the primary patent and literature searches. This level is especially useful where:
- important patents use unexpected terminology;
- ownership has changed;
- an inventor has continued the same chemical series in another family;
- related compounds were published scientifically; or
- a critical family was not retrieved by the original structure query.
It provides assurance that the results don’t hinge on a single query or a single database.
Best Practice: Conduct exact, substructure, and similarity searches and cross-check between databases, but have an expert review the search results as they are reported for each claim.
What Are the Challenges in Conducting a Chemical Structure Search?
- Markush complexity. Claims use variables such as R¹, R², Ar, Het, or X to represent a vast number of compounds, like “R-groups” refer to substituents attached to the core structure; “Het” for a heteroaryl or heterocyclic group. The challenge grows when variables are nested. For example, R² = –NH–Ar, where Ar is itself defined through several alternatives. A single structure query may not reveal whether your molecule falls inside a generic formula, and coverage is not the same as disclosure.
- Inconsistent database encoding. Different patent databases encode generic structures differently, so the same query can return inconsistent results across platforms. Complex Markush claims, ambiguous structural variables, database incompatibility, and incomplete indexing are major causes of false negatives, which is why automated searching alone may not capture the full picture.
- Large or complex molecules. Big structures need to be broken into meaningful sub-units, or the search returns noise or misses relevant art.
- Naming and representation gaps. The same molecule appears under systematic names, code names, trade names, and drawings, and some structures sit in images that text search cannot read.
- Fragmented databases. No single source covers every jurisdiction or document type, so coverage depends on combining the right paid and free resources.
- Stereochemistry, salts, and forms. Ignoring these can mean missing relevant claims or burying the useful ones.
- Legal status and territory. A hit means little until you confirm whether the patent is in force where you operate.
- Judgment calls. Deciding whether a claim covers your compound takes a chemist and a patent analyst reading the claims together.
Best Practice: Never rely on a single query or a single database. Test whether the results can be expanded before treating them as complete.
What Is the Best Time to Conduct a Chemical Structure Search?
As with an FTO search, the best time is before a major commitment of money, legal exposure, or reputation. The key triggers are:
- During lead identification or optimisation, when you can still adjust a substituent to design around a blocking claim.
- Before filing a patent application, to confirm novelty and shape the claims.
- Before clinical investment or scale-up, when costs and risk rise sharply.
- Before launch or entering a new market, since patent rights are territorial.
- Before licensing, partnerships, or M&A, as part of IP due diligence.
- When facing a competitor’s patent, to look for invalidating prior art.
- Ahead of patent expiry or generic entry, to plan lifecycle strategy.
If you can run only one search, run it before the commitment that is hardest to reverse. If you can run two, add one early in development, while there is still room to redesign.
How Is AI Redefining Chemical Structure Search?
AI is transforming the way we work in four practical ways:
- Reading structures in images and text. New models are able to identify molecules in patent drawings and scan documents that previous tools could not.
- Smarter similarity. Machine-learning models reveal analogues that conventional fingerprint matching overlooks.
- Markush handling. AI-assisted tools can help check whether a compound falls within a generic formula or not (something that used to be mostly manual).
- Speed and prioritisation. Large result sets can be ranked and clustered so experts spend their time on the claims that matter.
AI does not replace judgment. The models can get context wrong and sometimes can’t understand whether the claims are enforceable or whether the disclosure is enabling, and they still struggle with nested variables and poor database indexing. AI-assisted retrieval reviewed by a subject-matter expert delivers the strongest results, the hybrid approach we have adopted at Ingenious e-Brain.
How Does Ingenious e-Brain Assist Companies Seeking Chemical Structure Search?
We bring together chemistry, patent search, and expert-level attorney judgment in a single workflow. We break down the structure, use AI to perform keyword and multi-level structural searches across platforms, confirm findings via reassurance searching, and draw insights claim by claim. The result is not just a list of patent references, but a view of where the molecule is positioned in the patent ecosystem, what rights merit deeper study, and where there’s still freedom to invent.
For the right question, we ask more than “Was the exact molecule found?”:
- Has the scaffold been disclosed?
- Are related analogues already known?
- Could a broader Markush claim cover the chemical space?
- Are hidden structural variants being missed by automated searching?
Final Thoughts
A chemical structure search is not a one-time task. As recent cases show, the difference between a protected product and a blocked one can come down to a single compound and the earlier patents around it. Run the search early, repeat it at key triggers, and pair AI-driven retrieval with expert review so that you protect your inventions and decide with confidence.
If you are working through one of these triggers, Ingenious e-Brain’s IP search team can help you build a clear, defensible position. Talk to our experts at contact@iebrain.com.
Frequently Asked Questions
1. What is a chemical structure search?
Chemical structure search is a patent and literature search based on a molecule’s structure rather than its name, covering exact matches, substructures, analogues, and Markush claims.
2. Why not just search by compound name?
Patents often claim compounds in a way that does not use a specific name. The same molecule can appear under different names or only in a drawing. So, a chemical structure search is more reliable.
3. Why can an exact chemical structure search be misleading?
An exact match in a chemical structure search shows only that a particular compound was not named. A broad Markush formula, a related scaffold, or a close analogue series can still reveal or cover the chemical concept. Thus, an exact chemical structure search may still miss relevant information.
4. What is reassurance searching?
It is a validation layer that uses two-level backward and forward citation analysis and further searches around relevant assignees, inventors, and authors to confirm that no key family was missed.
5. How often should I repeat a search?
The search should be repeated at each major trigger, treating anything older than 12 to 18 months as a candidate for refresh.
