Validation Module
The validation module checks ¹H and ¹³C assignments of a structure against nmrshiftdb2 quickcheck HOSE-code predictions. It is executed by nmr-cli (validate-assignments) inside the nmr-converter container.
Base path: /latest/validate
The prediction is a reference, not the truth: a poor fit flags assignments for a closer look, and the author keeps the final word.
Endpoints
POST /assignments
Request body:
{
"structure": { "molfile": "\n Mnova...\nM END", "source": "mnova" },
"conditions": { "solvent": "CDCl3" },
"assignments": [
{ "nucleus": "13C", "atoms": [1, 3], "label": "C-2, C-6", "shift": 106.65 },
{ "nucleus": "13C", "atoms": [5], "label": "C-4", "shift": 143.52 },
{ "nucleus": "1H", "atoms": [1, 3], "label": "H-2, H-6", "shift": 7.12, "n_h": 2 },
{ "nucleus": "1H", "atoms": [9], "label": "H-5'a", "shift": 2.31, "diastereotopic": "a" }
],
"unassigned_peaks": [{ "nucleus": "13C", "shift": 77.16, "kind": "solvent" }]
}| Field | Notes |
|---|---|
structure.molfile | V2000. Explicit H atoms are allowed and are stripped before the servlet call. |
assignments[].atoms | 1-based molfile indices. For ¹H: the carrying heavy atom, an explicit H, or an index past the last molfile atom for an implicit H numbered heavy atom by heavy atom (openchemlib addImplicitHydrogens, as NMRium does). Equivalent atoms share one entry. |
assignments[].label | Author label. "C-2, C-6" with two atoms gives one report row per atom. |
assignments[].diastereotopic | Marks the two protons of a CH₂; the pair is compared by its mean. |
unassigned_peaks | unknown peaks join the structure fit; solvent and impurity peaks are ignored. |
options.fallback_tolerances | Per nucleus {ok, fail} in ppm (defaults: ¹³C 3/6, ¹H 0.3/0.6). |
Response: a report with two layers and a combined verdict.
| Key | Question it answers |
|---|---|
reports["13C"], reports["1H"] | How well do the assigned shifts fit, atom by atom? Mark 1–10, accept/revise/reject, penalties, per-atom deviation, HOSE spheres and codes, in_database_likely. Laid out like the nmrshiftdb2 quality report, but on the author's assignments: each atom carries its assigned shift and the status of its assignment_check row, so both layers always agree. |
assignment_check | Are the shifts on the right atoms? Status per assignment (ok, review, fail, not_assessable), swap suggestions, equivalence violations, missing signals, solvent peaks, proton counts, referencing offset. |
verdict | accept, review, reject or not_assessable; ¹³C drives it. |
adjustments | Shifts nudged by ±0.001 ppm so the servlet keeps distinct signals with identical values. |
cached | Identical requests are served from a 24 h in-memory cache. |
Scoring
nmrshiftdb2 does not publish its mark formula. The mark is an approximation (0.5 points per ppm mean deviation, 2 per red or missing atom, 1 per yellow atom, halved for predictions with at most 2 spheres) and is flagged with mark_is_approximate. A nucleus with a red atom is at best revise, whatever its mark. Predictions with fewer than 4 HOSE spheres can lead to review, never to fail.
Errors:
| Status | Meaning |
|---|---|
| 422 | Invalid request, unsupported molfile or unknown atom indices |
| 408 | Validation timed out |
| 503 | nmrshiftdb2 quickcheck unavailable, retry later |
| 500 | Docker or nmr-converter not available |
CLI
cat assignment-set.json | nmr-cli validate-assignmentsExit codes: 2 invalid input, 3 quickcheck unavailable.
Related modules
- Prediction — nmrshift engine used as the reference
- Spectra — parse experimental data for validation input