Effect of "small" size libraries on the final result. The predictive pdf and the Bayes Error Rate E of both tumoral (T) and normal (N) classes for examples tags contained in the three important prototype cases described in text are shown. The 'x' and 'o' marks represent observed abundances in each tumoral and normal libraries. Frame a) shows the result using all libraries. The "small" libraries (size < 50,000) are highlighted with their counts over library size. Frame b) shows results excluding those "small" libraries. It is clear that results are pretty much the same and that "small" libraries are not (necessarily) outliers of the sampling.