nmrtoolbox

Introduction

This is a simple utility that provides modules for working with NMRPipe peak tables and performing a receiver operator characteristic (ROC) analysis to quantify the quality of the "recovered" peaks relative to a control set of "synthetic" peaks. The modules in this package are as follows:

• nmrtoolbox.peak: classes for reading in peak tables (currently supports both synthetically generated and recovered peak tables from NMRPipe)
• nmrtoolbox.roc: perform receiver operator characteristic (ROC) analysis of a recovered peak table relative to a synthetic peak table
• nmrtoolbox.mask: define regions of a spectrum that contain signal or are empty
• nmrtoolbox.util: various supporting utilities used by other modules
• nmrtoolbox.strip_plot: generate color coded strip plots that correspond to the true positive and false positive peaks classified by an ROC analysis

Applications

Example #1 - Formal Workflow

The tools in this package are utilized by the NUScon software package. You can access NUScon on the NMRbox platform (free for academic, government, and non-profit users). Running nuscon -h will provide instructions on how to run the NUScon evaluation workflow, which directly utilizes the tools presented here in the nmrtoolbox package.

Example #2 - Kick the Tires

from nmrtoolbox.roc import roc
from nmrtoolbox.strip_plot import strip_plot

# perform ROC analysis and specify filtering criteria
my_roc = roc(
    recPeaks=<file-recovered.tab>,
    synPeaks=<file-synthetic.tab>,
    cluster_type=1,
    chi2prob=.75,
    vol_height_mismatch=True,
    maxLW_percent_SW=0.25,
)

# show and plot results
my_roc.print_stats()
my_roc.plot_roc()

# generate strip plots
strip_plot(
    recSpectrum=<file-recovered-spectrum.ft3>,
    sumSpectrum=<file-sum-spectrum.ft3>,
    empSpectrum=<file-empirical-spectrum.ft3>,
    empPeaks=<file-empirical.tab>,
    roc_obj=my_roc,
)

The roc function supports the following filter criteria:

• number
• height
• abs_height
• roi_list
• index
• cluster_type
• mask_file
• chi2prob
• outlier
• vol_height_mismatch
• maxLW_percent_SW

Note: Filtering by mask requires the external use of NMRPipe to generate a mask file indicating where the spectrum is empty. This binary data is converted by Connjur Spectrum Translator into a "tabular" format file (i.e. plain text) which is then read in by nmrtoolbox.mask.

The strip plot function shown above will make pairwise comparisons among the following 3 types of input spectra (all of which are fundamental to the NUScon workflow):

• empSpectrum: empirical spectrum, this reference is typically obtained by FT processing of a uniformly collected experiment
• sumSpectrum: sum spectrum, this control is typically obtained by FT processing the uniformly sampled time domain data of the empirical data augmented with synthetic peaks
• recSpectrum: recovered spectrum, typically obtained by processing a nonuniformly sampled version of the empirical time data augmented with the synthetic time data

In addition, the corresponding peak tables are also accepted as inputs:

• empPeaks: empirical peaks, peak table from the empirical spectrum
• synPeaks: synthetic peaks, peak table of just the synthetic peaks used to build the sum spectrum
• recPeaks: recovered peaks, peak table from the NUS reconstruction of the synthetic and empirical data

Example data from NUScon archive is available on NMRbox at '/NUScon/archive'

Changelog

v11

• major upgrades to strip_plot module
• PeakTable and Spectrum classes use enumerated types to define valid input formats
• PeakTable and Spectrum classes offer .read() methods to handle multiple input types
• rename roc input parameters to be consistent with PeakTable and Spectrum classes
• universal change to variable naming: "injected" is no longer used. "synthetic" refers to only the synthetic peaks/spectra and "sum" refers to "empirical" + "synthetic" peaks/spectra

v10

• major change: ROC now sorts peaks by absolute value of intensity (previously used intensity from largest positive down to largest negative)
• new module added for generating strip plots
• add more options for peak table filtering

v9

• change in internal data model for storing metadata in Peak, PeakTable, Mask, ROC, and ROI classes
• allow roc class to accept Mask object (not just mask file)
• approximate maximum LW for injected peaks from X1/X3, etc. parameters in the injected peak table
• function to write NMRPipe peak table to file

v8

• change to MIT license
• box_radius for mask filtering is multidimensional
• improved input options for setting carrier frequency
• axis labels used to verify compatibility of Peak, Mask, ROI, and ROC objects

v7

• addition of roc module
• addition of mask module

v6

• rename package as nmrtoolbox
• use subclasses to handle NMRPipe peak tables coming from genSimTab or from the peak picker

v5

• new Axis class for containing metadata from peak table header