Skip to content

Processing & reconstruction

Spectral processing

Take raw time-domain data from the spectrometer to a phased, baseline-corrected, referenced spectrum in the browser.

Processing is where most analysis errors are actually made, and where they are hardest to see later. Window function, zero-filling, phase and baseline all propagate silently into every peak intensity you fit downstream, so a relaxation rate can be wrong by more than its error bar because of a decision made before the spectrum was ever displayed.

We run processing as an explicit, recorded pipeline rather than a script someone kept on a lab machine. Each stage — solvent suppression, apodisation, zero-fill, Fourier transform, phase, baseline, extraction, referencing — is a parameterised step, and the whole chain is stored with the spectrum it produced. Reprocessing with one changed parameter is a new run with its own provenance, not an overwrite.

Our nmrglue-based engine reproduces NMRPipe semantics stage by stage. We validate it against NMRPipe at every intermediate point rather than only on the final spectrum: on FID-Net's published 700 MHz carbon-proton dataset the two agree to float32 precision, with a correlation of 1.00000000 and a relative RMS deviation below 5e-06.

What it needs

  • Bruker, Varian/Agilent or NMRPipe time-domain data
  • Acquisition parameters (read automatically)

What you get

  • Processed 1D, 2D or pseudo-3D spectrum
  • The full processing chain, stored and re-runnable

What does the work

  • nmrglue

    Open source, redistributed under its licenceBSD-3-Clause

    Time-domain I/O and the processing kernel

  • pipeglue

    Built in-house

    Our NMRPipe-equivalent processing chain, validated stage by stage

Related in Processing & reconstruction