PNNL-13501 - Pacific Northwest National Laboratory

Analysis of High-Volume, Hyper-Dimensional Mixed Data: NMR Spectra and Confocal
Image Ensembles
Study Control Number: PN00009/1416
Don Simone Daly
New research tools for advanced biological studies generate highly complex, large volume, and multi-dimensional data.
Often, these datasets represent mixed media, such as spectra, picture images, and scalars.
Project Description
Scientific research often generates complex, large
volume, or hyper-dimensional datasets. Analyzing
complex, multi-dimensional datasets can be highly
challenging. The object of this project was to provide the
tools, techniques, and expertise to analyze ensembles of
spectra and images as an ensemble. Making the step from
characteristics of a single object (such as an NMR
spectrum) to the characteristics of the ensemble (i.e., an
NMR spectral time series) provides new opportunities to
unravel complex associations. We investigated the tools,
techniques, and expertise necessary for the analysis of
mixed sets of NMR spectra and optical images.
Results and Accomplishments
The primary accomplishments of this project are 1) the
introduction of a new perspective from which to view the
collection and analysis of NMR and optical datasets, and
2) the development of new analysis methods made
possible by this perspective. Our new perspective
emphasizes the data ensemble over the individual
measurements. The new methods are geared to the
collection and analysis of the ensemble.
We reexamined the NMR phenomenon and its subsequent
measurement. Our examination produced a stochastic
model of this spectral measurement that is true to the
physics and includes the measuring uncertainties. Our
model suggested a graphical method to view an NMR
quadrature spectrum and an NMR spectral ensemble.
Our stochastic model describes an NMR-quadrature freeinduction-decay
measurement as a complex-valued time
series. This model suggested that a phase-space graph
plotting the real component of the measured time series
against the imaginary component would provide not only
information about the phenomenon under study, but also
about the effects of measuring this phenomenon using
NMR (Figure 1).
Figure 1. Phase space representation of a complex-valued
NMR free induction decay of a sample featuring two
dominant NM environments
A quick assessment of an ensemble of NMR
measurements is gained by using the phase-space graph as
a glyph to represent one NMR free induction decay on a
graph of the ensemble (Figure 2). In this figure, the
variation in the glyph over the underlying image hints at
the varying spatial distribution of chemicals within the
sample. The corner glyphs, where no signal is present,
indicate that the measuring noise is random and
uncorrelated between the two components of the
complex-valued free induction decay measurements.
Under this project, APEX, a set of spectral feature
extraction algorithms developed for MALDI-mass
spectrometry was extended to extract features from an
NMR spectrum. The extended algorithms automatically
extract objective, robust estimates of peak locations,
heights, and other peak characteristics with associated
standard errors. The collection of estimates for each
spectrum was then analyzed further.
We also explored the analysis of original and extracted
spectral collections using spectral mixing models and
spectral unmixing algorithms. Under a spectral mixing
Statistics 445