PLS-DA is a regression technique which maximizes the separation
between pre-defined classes. The aim is to predict the values of
a group of variables X (dependent variables) from a set of variables
Y (explanatory variables). In our case, Y represents quantitative
variables with “band volume” (surface of the band in pixel multiplied
by the intensity level of each pixel of the band value measured
by ImageQuant TL software) and X represents qualitative variables,
that is to say the type of farming (organic, conventional or sustainable).
In our study, binary classification models were developed
(for example 1 for organic and 0 for conventional) and the
belonging to one of the classes was predicted by PLS-DA according
“bacterial band intensity” value. To optimize the number of PLS-DA
latent components (LV), the percentage of correct classifications
(sensitivity and specificity) obtained was validated.