g p

)

[

y

,

]

,

y

t to pay the attention to search for an optimal regularisation

so as to guarantee an optimal baseline to be estimated for the

signal discovery.

to determine an optimal value for the regularisation constant is

al. The value of the regularisation constant must be enumerated

to determine at which point, an optimal baseline model can be

d for a spectrum. Figure 5.4 shows how the fidelity measurement

moothness measurement vary when the regularisation constant

As shown in Figure 5.3(c), when the regularisation constant

too large, an estimated baseline tends to be more and more

resulting in a great error or fidelity. This is also shown in Figure

at when the regularisation constant is too small, a curve fits to

ail of the intensities in a spectrum. Therefore a baseline estimated

will not be sufficiently smooth.

e impact of the regularisation constant on the error (fidelity) and the smoothness

odel.

ical, but less focused issue is how to handle multiple spectra, i.e.,

data with more than one replicate. Most baseline estimation

s have not yet dealt with this issue very well. A baseline

n model constructed for each single spectrum will generate a peak

after a baseline has been estimated and removed from each

Figure 5.5 shows a spectra data set with four replicates. Based on

r replicates, a baseline estimation algorithm can be applied to

a baseline for each spectrum. Based on the estimated baseline, a