Table 3.

Localisation accuracies for the fixed set of protein locations.

			I		II
feature set	reduction method	#features	$\overline{ACC}$	ACC	$\overline{ACC}$	ACC
$\mathcal{A}$	-	73	0.777	0.799	0.790	0.828
ℬ	-	73	0.818	0.833	0.806	0.835
$\mathcal{A}$	SDA	17	0.860	0.873	0.873	0.887
ℬ	SDA	15	0.878	0.891	0.883	0.892

Although the SFAMs were trained using both manually and automatically obtained protein distribution patterns, the evaluation is performed separately. Here, I denotes the accuracies regarding the manually acquired patterns and II the results with respect to the automatically generated samples. The results indicate an advantage of feature set ℬ in comparison to feature set $\mathcal{A}$. Moreover, the feature reduction leads to a significant improvement of the protein localisation, independent from the feature set applied.