Chemistry Reference
In-Depth Information
2. Methodology
The mass spectral data selected for multivariate analysis represented a
suite of 30 microlayer and bulk surface seawater surfactant samples (frac-
tion F1; Frew et al. (2006)) including seven from waters off Monterey and
Santa Barbara, California and 23 collected along a transect from Delaware
Bay on the U. S. east coast to the Sargasso Sea. Sample extracts were ana-
lysed in triplicate by desorption-electron ionisation (DEI) mass spectro-
metry (Boon 1992; Frew et al. 2006). Individual DEI mass scans were
summed over the full desorption/pyrolysis interval, reduced to integer
masses, and averaged for processing by multivariate analysis. Elasticities
were estimated quasi-statically from surface pressure-area (3/
A
) isotherm
measurements in a KSV 2200 Langmuir film balance (Frew et al. 2006).
The elasticity data were subsampled at fixed surface pressure intervals (0.5
mN m
-1
) for comparison with the results of the multivariate analysis.
Linear discriminant analysis (LDA) was applied to the resulting 90 mass
spectra using a MATLAB-based (MATLAB™, The Mathworks, Inc., Na-
tick, Massachusetts) chemometrics routine 'Chemometricks' (developed at
FOM-AMOLF, Amsterdam; www.amolf.nl). This LDA routine was im-
plemented using a two-step principal component analysis (Hoogerbruge et
al. 1983) that reduced the number of variables relative to the number of
classes. Classes were defined in two ways: by allowing each set of sample
replicates to represent one class, yielding 30 classes, or by arbitrarily bin-
ning samples with similar elasticity values at 3 = 6.5 mN m
-1
. The seven
classes resulting from the latter method are listed in Table 1. Discriminant
functions were then developed that minimised and maximised the within-
class and between-class variance (Windig et al. 1983) respectively, and the
discriminant function scores for each spectrum were evaluated. The possi-
ble relationship between the mass spectral patterns and elasticity was then
evaluated by computing the correlation matrix for the static surface elasti-
cities measured for each surface film at each surface pressure interval and
the corresponding discriminant scores, resulting in linear relations between
elasticity and discriminant score for each surface pressure interval. A
quantitative description of microlayer films covering different dynamic
and physical conditions in a variety of geographic and productivity re-
gimes was thus developed. In this initial report, LDA results are presented
for fraction F1 extracts only. Similar LDA analyses of fractions F2, F3
and the SepPak isolates from this sample suite are in progress and will be
reported elsewhere.
