Skip to content

R/S descriptor builds the other configuration at one centre in some polycyclic names (CIP ranking differs from centres) #349

Open

Description

@Kohulan

Hi Daniel,

For a set of polycyclic names, OPSIN (2.9.0 and master 42e87ac6efb1) builds the stereoisomer that has the opposite
configuration at exactly one stereocentre. At that centre, SiMolecule/centres and RDKit's new CIP labeler
(rdCIPLabeler, a port of centres) give the descriptor written in the name; when they label OPSIN's output, they give
the other letter. Legacy RDKit agrees with them in most cases. So the difference seems to be in how the substituents of
that centre are ranked, not in the name.

Example 1 (centre 8; the tie between the cinnoline C-8a and the pyrimidin-4-yl carbon is in mancude rings, P-92.1.4.4):

(5R,8S)-8-(2-chloropyrimidin-4-yl)-3-(2,6-difluorophenyl)-5,6,7,8-tetrahydro-5,8-methanocinnoline
expected: Fc1cccc(F)c1-c1cc2c(nn1)[C@@]1(c3ccnc(Cl)n3)CC[C@@H]2C1           PAHBBFFRTVGGSE-DGIBIBHMSA-N
OPSIN:    ClC1=NC=CC(=N1)[C@]12CC[C@@H](C=3C=C(N=NC13)C1=C(C=CC=C1F)F)C2    PAHBBFFRTVGGSE-GIGQVBGESA-N

Example 2 (centre 2, no aromatic ring; C3 > C1 > C4, and C1 beats C4 only at sphere 4 through the ring-closure duplicate
of C2 on the C4 branch):

(1S,2R,3R,5R,6R,7S)-3-(ethoxycarbonyl)-7-(methoxycarbonyl)tricyclo[3.2.2.0^2,4]non-8-ene-6-carboxylic acid
expected: CCOC(=O)[C@@H]1C2[C@@H]1[C@@H]1C=C[C@H]2[C@@H](C(=O)O)[C@H]1C(=O)OC    FVZBYJXSSLHZGY-JSYYRQLRSA-N
OPSIN:    C(C)OC(=O)[C@H]1[C@@H]2[C@H]3[C@@H]([C@@H]([C@@H](C12)C=C3)C(=O)O)C(=O)OC FVZBYJXSSLHZGY-FAGYLJKTSA-N

Example 3 (centre 6):

(3aS,6S,6aS,9aR,10aS,10bR)-8-(bromomethyl)-6a,9a-dichloro-6-(4-hydroxyphenyl)-2-methyl-3a,4,6,6a,9a,10,10a,10b-octahydroisoindolo[5,6-e]isoindole-1,3,7,9(2H,8H)-tetrone
expected: CN1C(=O)[C@H]2[C@H](CC=C3[C@H]2C[C@@]2(Cl)C(=O)N(CBr)C(=O)[C@@]2(Cl)[C@H]3c2ccc(O)cc2)C1=O    CNYQHBKBWZBHGB-KJBJECQHSA-N
OPSIN:    BrCN1C([C@]2([C@@H](C=3[C@H]([C@H]4C(N(C([C@H]4CC3)=O)C)=O)C[C@]2(C1=O)Cl)C1=CC=C(C=C1)O)Cl)=O    CNYQHBKBWZBHGB-COCOWISSSA-N

In a sample of 1,899 polycyclic structures, 142 names show this pattern (one centre; centres and rdCIPLabeler agree
with each other). In #217 you mentioned possibly integrating centres; these could serve as test cases for that. Happy to
send the full list.

Activity

  1. dan2097 commented on Nov 2, 2026

    @dan2097
    Owner

    This appears to be due to a lack of support for atomic number averaging in mancude rings in OPSIN's CIP implementation. I'm not immediately sure whether it makes sense to implement this functionality into the existing code or dump the code entirely and simply use centres in all cases.
    @Kohulan Would it be possible to email me the complete list.

  2. self-assigned this
    on Nov 2, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

Labels

bugSomething isn't workingenhancementNew feature or request

Projects

No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions