My inactive enantiomer came back as a hit

Your screen flags a compound your assay says is inactive. You check the structure in the vendor’s catalogue and it is the right molecule. You check the key and it matches. What you cannot see is that the key you matched on stopped distinguishing the two enantiomers somewhere upstream, so the hit is the other one — and the one that works is in your set too, under the same name.

What actually happened

Standardising a compound set usually involves picking one tautomer per tautomeric form, so that warfarin’s keto and enol forms stop being two compounds. That is correct and recommended. RDKit does it with TautomerEnumerator().Canonicalize.

It also discards stereochemistry at an sp3 centre inside a tautomeric system. A carboxylic acid’s alpha-carbon is inside one. So:


ibuprofen (S)   C[C@@H](c1ccc(CC(C)C)cc1)C(=O)O   HEFNNWSXXWATRW-JTQLQIEISA-N
ibuprofen (R)   C[C@H](c1ccc(CC(C)C)cc1)C(=O)O    HEFNNWSXXWATRW-SNVBAGLBSA-N

both, after tautomer normalisation                 HEFNNWSXXWATRW-UHFFFAOYSA-N

Two keys become one. The JTQLQIEISA and SNVBAGLBSA blocks are the stereo layer, and after normalisation there is no stereo layer: UHFFFAOYSA is what an InChIKey says when there is nothing to say.

This is RDKit behaving as documented, not a bug in it. Nothing was corrupted and no file is malformed. What changed is what a name in the file means.

Why this is the dangerous direction

The other standardisation failure makes an overlap too small: two sets disagree about one substance, so they fail to match where they should. You notice an empty answer eventually.

This one makes an overlap too large, and specifically wrong. Two substances collapse onto one name, so a join matches a pair that is not a pair. For a racemic library against a stereo-resolved hit list, every single compound matches something — and roughly half of those matches are the enantiomer you did not mean.

For a drug set that is not a technicality. Two enantiomers are routinely one active compound and one inactive one, and sometimes one drug and one toxin. Thalidomide is the example everybody knows; ibuprofen, levodopa, esomeprazole and dozens of marketed single-enantiomer drugs are the reason anybody resolves a racemate in the first place.

Why nothing in the file says so

An InChIKey carries its stereo layer in the key, so the information is not hidden — it is absent. HEFNNWSXXWATRW-UHFFFAOYSA-N is a complete, well-formed key for a molecule with no defined stereochemistry, and a column of those is indistinguishable from a column describing genuinely achiral compounds. There is nowhere in a table of keys to record that stereochemistry was lost rather than never present.

And the structure column, if the file has one, agrees with the key: the canonical tautomer’s SMILES is CC(C)Cc1ccc(C(C)C(=O)O)cc1, with the [C@@H] gone. So the file is internally consistent and externally wrong.

Why the design was wrong about this

Worth recording, because it was. §4.4 of the v0.2 design specified the standardisation state as “salt-stripped, tautomer-normalised, charge-neutralised, stereo kept” — four things you ask for, with stereo among them.

Stereo is not something you keep. It is something tautomer normalisation takes, and no option in the design would have brought it back. The measurement that found this is on the architecture page; the project’s habit is to record where a specification was wrong rather than quietly correct it.

So representation records stereo-discarded as an operation that was applied — observed, not requested — and the standardiser checks what it did against what it was asked for.

What Ligase says instead

Two libraries alike in every other respect, one of which lost its stereochemistry:

DimensionErrorin stepscoreligase check — nothing ran

library.representation=InChIKey, salt-stripped and charge-neutralisedfrom the representation line the standardiser wrote: crates/ligase-core/tests/fixtures/molecules_library.tsv

hits.representation=InChIKey, salt-stripped, charge-neutralised and stereo-discardedfrom the representation line the standardiser wrote: crates/ligase-core/tests/fixtures/molecules_library_stereo_lost.tsv

hits also lists stereo-discarded.

These must match. Were the two sets standardised differently? Add a `standardize` step; molecules are never re-keyed implicitly.

One line names the operation, and it is the only part worth reading: the two value lines are otherwise identical.

And the check that stops it at the source

A set only reaches that state by being standardised, so the standardiser refuses to produce one that does not admit what it did:


clean = lg.chem.standardize(library, to="inchikey; tautomer-normalised")


ValueError: standardising discarded stereochemistry on 418 of 2314 molecules,
and the state you asked for does not say so. RDKit removes stereochemistry at
an sp3 centre inside a tautomeric system, and a carboxylic acid's
alpha-carbon is inside one -- so ibuprofen's two enantiomers both become
HEFNNWSXXWATRW-UHFFFAOYSA-N. Add `stereo-discarded` to `to=`, so this set
refuses to join one that still knows which enantiomer it holds, or drop
`tautomer-normalised` to keep the difference.

You get the count before you choose, and the two choices are real ones:

Add stereo-discarded. The set is then honest about what it has lost, and will refuse to join a set that kept the distinction — which is this page’s diagnostic, arriving before the join rather than after the conclusion.
Drop tautomer-normalised. You keep the enantiomers apart and lose the tautomer matching.

There is no third option that gives you both, because RDKit does not offer one.

Where it bites hardest

A racemic library screened against a resolved hit list. Every compound matches, half of the matches are the wrong enantiomer, and the enrichment figure is meaningless rather than merely noisy.

A chiral QSAR model. Two enantiomers with different measured activities become one training example with two labels. The model is being asked to learn a contradiction, and will report that this endpoint is hard to predict.

Deduplicating a catalogue. A racemate and its resolved forms count as one compound, so “unique compounds” is an undercount and nothing says by how much.