Skip to content

Request for feedback: validate new Enhanced Stereochemistry InChIs #286

Description

@cm-beilstein

Request for feedback: Enhanced Stereochemistry InChIs

We are working on support for enhanced stereochemistry (-EnhancedStereochemistry) and are looking for community feedback on the InChIs it produces.

Review page: https://cm-beilstein.github.io/InChI-enh-Stereo/

The page lists 107 test molecules in four groups:

  1. Enhanced stereochemistry (ABS/OR/AND): absolute, relative and racemic configurations
  2. Allenes: axial stereochemistry
  3. Double-bond E/Z: geometric isomers in stereo collections
  4. Atropisomers: stereoisomers caused by restricted rotation

Each entry shows:

  • the structure (SVG with canonical atom numbering)
  • a description of the stereo feature being tested
  • the old InChI (current dev)
  • the new InChI with -EnhancedStereochemistry, with the changed tail segments highlighted in red

How you can help

If you are interested, please look through the structures and check whether the new InChIs are correct. We especially want to hear about:

  • InChIs that are wrong or don't match the drawn structure or the intended stereo collection (abs, or<n>, and<n>)
  • edge cases that behave unexpectedly or are missing
  • comments on the notation itself (readability, consistency)

Please leave your feedback as a comment on this issue and give the ID number of each molecule you mean.

Development branch: feature/enh-stereo-hardening

Activity

  1. linked a pull request that will close this issueFeature/enh stereo hardening #251on Oct 15, 2026
  2. pechersky commented on Oct 16, 2026

    @pechersky

    Hi, not sure why #277 is linked here, that one has no enhanced stereo, and the double bond isn't in a stereo collection.

  3. nbehrnd commented on Oct 16, 2026

    @nbehrnd

    If Ketcher as provided on the landing page (version 3.17.0) was used to prepare the atropisomers, there might be a round-trip problem.

    Thanks to the Jmol project, I'm aware of a CIP validation suite (Robert Hanson et al., doi.org/10.1021/acs.jcim.8b00324, GH repo, synopsis). Thus I redrew from scratch the entry 55

    Image

    in the editor

    Image

    to then export the extended SMILES string (floppy disk icon) as

    C1(C(Br)=CC=CC=1C(O)=O)C1C(Br)=CC=CC=1C(O)=O |wU:10.11,(5.55,-4.35,;5.05,-3.48,;5.54,-2.61,;4.05,-3.49,;3.55,-4.35,;4.05,-5.22,;5.05,-5.22,;5.05,-6.22,;5.92,-6.72,;4.18,-6.72,;6.55,-4.35,;7.05,-5.22,;6.56,-6.09,;8.05,-5.21,;8.55,-4.35,;8.05,-3.48,;7.05,-3.48,;7.05,-2.48,;6.18,-1.98,;7.92,-1.98,)|
    

    If pasted to CDK depict, the representation substitutes the solid wedge by the dashed one:

    Image

    which on first view looks like the enantiomer. Not sure where the origin of this issue is, I filed a ticket over at CDK Depict.

  4. johnmay commented on Oct 17, 2026

    @johnmay
    Contributor

    @nbehrnd put in as a MOLfile and it should work BTW

  5. johnmay commented on Oct 17, 2026

    @johnmay
    Contributor

    Wasn't clear from the review page or if there is a place I can test?

    Here are some things I would want to check.

    1. MDL Chiral Flag

    Is MDL chiral flag the same as a set of &1/&2?

    Accelrys/BIOVIA semantics are chiral flag = 0 is treated the same.

    Sorry easier to provide as CXSMILES but these should then all be the same:

    C[C@H](O)CN |r|
    C[C@H](O)CN |&1:1|
    C[C@H](O)CN |&2:1|
    

    2. Meso Compounds

    Are the flags correctly ignored on cases like this?

    O[C@H]1CC[C@H](O)CC1
    O[C@H]1CC[C@H](O)CC1 |r|
    O[C@H]1CC[C@H](O)CC1 |&1:1,4|
    O[C@H]1CC[C@H](O)CC1 |o1:1,4|
    

    And these

    O[C@H]1CC[C@H](OCC[C@H](N)C)CC1
    O[C@H]1CC[C@H](OCC[C@H](N)C)CC1 |&1:1,4|
    O[C@H]1CC[C@H](OCC[C@H](N)C)CC1 |o1:1,4|
    

    3. Mixtures

    Does it normalise mixtures of isomers?

    See slide 13 - https://www.nextmovesoftware.com/talks/Mayfield_BuildingOnSand_InChI_201708.pdf

    CC[C@H](O)N.CC[C@@H](O)N |Sg:c:0,1,2,3,4,Sg:c:5,6,7,8,9,Sg:mix:0,1,2,3,4,5,6,7,8,9,SgH:2:0.1|
    
    Image
  6. nbehrnd commented on Oct 21, 2026

    @nbehrnd

    @johnmay Yes I know, CDK depict can process one .mol at a time (while multiple SMILES strings at once) - at least when accessing the demo page with firefox 153.5.0esr as repackaged for Linux Debian.

  7. cm-beilstein commented on Oct 23, 2026

    @cm-beilstein
    CollaboratorAuthor

    Hi @johnmay,
    thanks for your feedback. I converted your CXSMILES to V3000 molfiles with RDKit (2026.09.1), mapping |r| to chiral flag 0, and then ran them on branch (feature/enh-stereo-hardening, 773c921) with -EnhancedStereochemistry. Only the stereo layers are shown below.

    Unfortunately, we have only partly integrated this in the web demo yet.

    1. Chiral flag
    C[C@H](O)CN |r|                          /t3-/m0/s1   (flag ignored by default)
    C[C@H](O)CN |r|            -SUCF    /t3-/s2 (also results in non-standard InChI)
    C[C@H](O)CN |&1:1|                    /t3-/s3
    C[C@H](O)CN |&2:1|                    /t3-/s3
    

    Chiral flag 0 is not treated as &1. By default InChI ignores the flag. With -SUCF, InChI has always read flag 0 as relative (/s2), following the original MDL definition and the ChemAxon r definition. Mapping it to racemic (/s3) as BIOVIA does would change that existing meaning.

    1. Meso compounds: that is a bug
    O[C@H]1CC[C@H](O)CC1                           /t5-,6-
    O[C@H]1CC[C@H](O)CC1 |r|                       /t5-,6-
    O[C@H]1CC[C@H](O)CC1 |&1:1,4|              /t5-,6-/s3
    O[C@H]1CC[C@H](O)CC1 |o1:1,4|               /t5-,6-/s2
    

    If mirroring the group gives the same molecule, ignore the group. I will fix this.

    O[C@H]1CC[C@H](OCC[C@H](N)C)CC1                    /t8-,9-,10-/m1/s1
    O[C@H]1CC[C@H](OCC[C@H](N)C)CC1 |&1:1,4|       /t8-,9-,10-/m1/s3(9,10)
    O[C@H]1CC[C@H](OCC[C@H](N)C)CC1 |o1:1,4|       /t8-,9-,10-/m1/s2(9,10)
    

    This is a little bit more complicated to fix.

    1. Mixtures: not normalised
    CC[C@H](O)N.CC[C@@H](O)N |Sg:mix...|   2C3H9NO/.../t2*3-/m10/s1
    CC[C@H](O)N |&1:2|                                     C3H9NO/.../t3-/s3
    

    InChI ignores Sgroups, so the explicit enantiomer pair stays a two-component InChI. Collapsing mixtures into & groups is outside the scope of this work. We can open a separate issue for it if there is interest.

    I hope this makes sense. If you have any questions or find other bugs, please let me know. Thanks!

    8. Wasn't clear from the review page or if there is a place I can test?
    Here are some things I would want to check.
    # 1. MDL Chiral Flag
    Is MDL chiral flag the same as a set of &1/&2?
    Accelrys/BIOVIA semantics are chiral flag = 0 is treated the same.
    Sorry easier to provide as CXSMILES but these should then all be the same:
    C[C@H](O)CN |r| C[C@H](O)CN |&1:1| C[C@H](O)CN |&2:1|

    # 2. Meso Compounds
    Are the flags correctly ignored on cases like this?
    ```
    O[C@H]1CC[C@H](O)CC1
    O[C@H]1CC[C@H](O)CC1 |r|
    O[C@H]1CC[C@H](O)CC1 |&1:1,4|
    O[C@H]1CC[C@H](O)CC1 |o1:1,4|
    ```
    
    
        
          
        
    
          
        
    
        
      
    And these
    ```
    O[C@H]1CC[C@H](OCC[C@H](N)C)CC1
    O[C@H]1CC[C@H](OCC[C@H](N)C)CC1 |&1:1,4|
    O[C@H]1CC[C@H](OCC[C@H](N)C)CC1 |o1:1,4|
    ```
    
    
        
          
        
    
          
        
    
        
      
    # 3. Mixtures
    Does it normalise mixtures of isomers?
    See slide 13 - https://www.nextmovesoftware.com/talks/Mayfield_BuildingOnSand_InChI_201708.pdf
    ```
    CC[C@H](O)N.CC[C@@H](O)N |Sg:c:0,1,2,3,4,Sg:c:5,6,7,8,9,Sg:mix:0,1,2,3,4,5,6,7,8,9,SgH:2:0.1|
    ```
    
    
        
          
        
    
          
        
    
        
      
    <img alt="Image" width="266" height="75" src="https://private-user-images.githubusercontent.com/983232/676585942-6637a141-fd71-4659-bfda-a6f3f87ec174.png?jwt=eyJ0eXAiOiJKV1QiLCJhbGciOiJIUzI1NiJ9.eyJpc3MiOiJnaXRodWIuY29tIiwiYXVkIjoicmF3LmdpdGh1YnVzZXJjb250ZW50LmNvbSIsImtleSI6ImtleTUiLCJleHAiOjE3OTI3NDA3NjgsIm5iZiI6MTc5Mjc0MDQ2OCwicGF0aCI6Ii85ODMyMzIvNjc2NTg1OTQyLTY2MzdhMTQxLWZkNzEtNDY1OS1iZmRhLWE2ZjNmODdlYzE3NC5wbmc_WC1BbXotQWxnb3JpdGhtPUFXUzQtSE1BQy1TSEEyNTYmWC1BbXotQ3JlZGVudGlhbD1BS0lBVkNPRFlMU0E1M1BRSzRaQSUyRjIwMjYxMDIzJTJGdXMtZWFzdC0xJTJGczMlMkZhd3M0X3JlcXVlc3QmWC1BbXotRGF0ZT0yMDI2MTAyM1QwNzI3NDhaJlgtQW16LUV4cGlyZXM9MzAwJlgtQW16LVNpZ25hdHVyZT1jYzkxYjcxNjE1ZDI5YTVmZTMzYjdkMDEzOTYwYzg5ZTI1ZGI1Yjk1M2M1YmY4NDRjMDczZDRkNzU3YTk4MzJhJlgtQW16LVNpZ25lZEhlYWRlcnM9aG9zdCZyZXNwb25zZS1jb250ZW50LXR5cGU9aW1hZ2UlMkZwbmcifQ.5o_W1405VUrocemdHaB6r6TBwbY8PeHdhQQwbLocPas">
    
  8. johnmay commented on Oct 23, 2026

    @johnmay
    Contributor

    @cm-beilstein Thanks for getting back. Yeah in general it's a can of worms. So sure you know (due to Gerd) but is was MDL -> Symyx -> (SciTegic ->) Accelrys -> BOVIA. My understand was not that the original flag means "relative" as in "this or that" but rather "relative as in the relative configuration is known" the structure it is either Racemic or Relative, this is why ChemAxon treat it essentially as something completely different.

    It is reasonable to have a business rule / flag which this normalizes this. Perhaps a flag "treat chiral flag off as Relative" or "treat chiral flag off as Racemic".

    Here is the relevant section from the valuable MDL/BIOVIA "Chemical Representation" manual:

    Image
  9. cm-beilstein commented on Oct 23, 2026

    @cm-beilstein
    CollaboratorAuthor

    @johnmay Ok, I understood that wrong. I also saw that there already is an open GHI #221 about it. :D

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions