Table 5.

Localisation accuracies regarding the automatically generated patterns.

			M		M+A
feature set	reduction method	#features	$\overline{ACC}$	ACC	$\overline{ACC}$	ACC
$\mathcal{A}$	-	73	0.782	0.816	0.790	0.828
ℬ	-	73	0.794	0.836	0.806	0.835
$\mathcal{A}$	SDA	17	0.867	0.883	0.873	0.887
ℬ	SDA	15	0.880	0.890	0.883	0.892

These results were obtained using only manually acquired protein distribution patterns (M) and both manually and automatically obtained patterns (M+A) for training. They correspond to column II of Table 3.