NMR Prediction

NMR Prediction

This page explains the optional NMR Shift Predictor plugin, its experimental-data lookup method, and the limits of its shifts and spectrum displays.

Install and run

NMR means nuclear magnetic resonance. ChemDraft's NMR Shift Predictor is an official plugin installed separately from the core app.

1. Choose Plugins → Add or Remove Plugins… and install the NMR Shift Predictor. See Installing and Updating Plugins for package review and updates.
2. Select the structure you want to investigate.
3. Choose Analyze → Predict ¹³C NMR Shifts or Analyze → Predict ¹H NMR Shifts (experimental).
4. Read the shift table together with its spectrum, method information, and warnings.

On macOS those menus are app-wide. On Windows they belong to the document window. The plugin's commands appear after installation. Predictions are stored as analysis results and flagged stale when the structure changes; rerun the prediction after editing.

The plugin declares selection.read, analysis.write, ui.menu, and ui.panel. It reads the selected structure and writes analysis, rather than obtaining a document-editing capability.

Where shifts come from

The shipped backend is the OCL-native provider. OpenChemLib derives HOSE codes: descriptions of progressively larger atom neighbourhoods. The predictor looks these up, deepest matching sphere first, in aggregated shift statistics derived from NMRShiftDB2 experimental assignments.

A database match is evidence from related environments, not a measurement of your molecule. The available median/mean statistic describes those matched assignments. Coverage depends on the chemistry represented in the database.

Some bounded, versioned additive rules can supply disclosed estimates or a second opinion where their tables apply. They do not turn unsupported chemistry into a generic carbon or alkyl environment. The panel distinguishes the HOSE result from increment-estimator information and reports coverage. A rule estimate stays labelled as an estimate.

An unmatched environment does not receive an invented shift. Partial predictions keep warnings such as NMR_NO_FRAGMENT_MATCH and NMR_PARTIAL_PREDICTION. Read coverage before interpreting a tidy spectrum as complete.

Read the display correctly

Label or quantity	What it means
Chemical shift	Predicted position in parts per million (ppm)
¹H multiplicity	A first-order estimate from bond topology
J coupling	A first-order estimate in hertz, not a measured coupling
Equivalent nuclei / nEquivalent	Predicted number of nuclei grouped into that resonance
Stick height	Equivalent nuclei, not experimental integration
Field and line shape	Simulation parameters where offered, rather than measured instrument conditions
Confidence tier	Qualitative evidence; not a calibrated probability

The ¹³C output is proton-decoupled. A first-order proton multiplet is a simplified model; strong coupling and second-order effects can require a full spin analysis. Keep the experimental label on the proton command when describing its capabilities.

Labile protons are exchangeable O–H, N–H, and S–H hydrogens. The desktop proton command omits them by default. The CLI has a separate --ignore-labile option and does not share that default: inspect the chosen interface's options before comparing results.

Equivalence and stereochemistry

Grouping follows constitutional equivalence, which can merge nuclei that are chemically different because of stereochemistry. Such nuclei are diastereotopic. Examples include two CH₂ hydrogens or certain pairs of methyl groups.

The predictor discloses detected cases instead of inventing separate shifts. Warnings can identify potentially diastereotopic hydrogens or methyls, merged stereo-nonequivalent classes, and likely second-order patterns. If a disclosure check cannot complete, that limitation is reported alongside available shifts.

Do not split a warned resonance by hand or interpret one predicted line as proof of experimental equivalence. Preserve the structure-to-atom mapping when explaining which environment a resonance represents.

Compare with experiment

Place predictions beside the experimental spectrum with a clear Predicted label. Keep the input structure, statistic, method, database provenance, and warnings. Do not move shifts or alter the structure merely to fit the trace.

Use honest tiers and thin-match warnings; no calibrated confidence percentages are supplied. Accuracy claims require the exact benchmarked corpus and checksum gate. This page makes no general accuracy claim.

Agents and data licensing

The CLI nmr and MCP predict_nmr use the external plugin. Headless loading additionally requires its directory to be listed in the user's trust file; setting an environment variable alone grants no trust. CLI spectra are stick spectra, without field or line-shape controls. See Agent Setup Recipes.

The derivative database carries the nmrshiftdb2 Database License, with ODbL-derived attribution and share-alike terms, separately from the code licence. Preserve database attribution when redistributing data or results as required by those terms. The data-bundling package must not be described simply as MIT.

See also

• Analysis and Properties
• Official Plugins
• Plugin Permissions
• Chemistry Guarantees