2.4.2 Model Development for Partial Least Squares Discriminant
Analysis
The UnscramblerX software, similar to the PCA model development, was
used to develop the PLS-DA models. Three PLS-DA models were developed to
discriminate between the three microorganisms. These models were
designed for binary classification with two classes: samples containing
specific microorganisms (y = 1) or those without microorganisms (y =
-1). The models were developed using 7 latent variables, cross-validated
results with 20 segments of random samples, and the scores and loadings
were calculated using the NIPALS algorithm. The threshold for
distinguishing between the two classes was zero (0). Several figures of
merit were used to evaluate the PLS-DA model performance: the percent of
correctly classified samples (%T), percent of incorrectly classified
samples (%F), sensitivity (SEN), specificity (ΞΎ), and Matthews
correlation coefficient (MCC). A value of negative one (-1) for MCC
indicates the perfect classification of the opposite variable, a value
of one (1) is the correct classification of the class, and zero (0)
indicates random classification.