Skip to main content
. 2008 Sep 1;9(Suppl 2):S12. doi: 10.1186/gb-2008-9-s2-s12

Table 4.

Comparative results for interaction pair extraction

Compared models Whole collection SwissProt only article collection
Precision Recall F1 score Precision Recall F1 score
Mean 0.1062 0.1858 0.1035 0.1160 0.2000 0.1127
Standard deviation 0.0945 0.1001 0.0761 0.1035 0.1062 0.0836
Median 0.0755 0.1961 0.0788 0.0808 0.2156 0.0842
Best reported in terms of F1 score [8] 0.3908 0.2970 0.2849 0.3893 0.3073 0.2885
Template-based method (threshold = 0.0) 0.1373 0.2905 0.1578 0.1566 0.3189 0.1784
Template-based method (threshold = 80.0) 0.2177 0.2651 0.2038 0.2434 0.2828 0.2247
Profile-based method 0.3096 0.2935 0.2623 0.3695 0.3268 0.3042

'Whole collection' means that all of the articles have been considered. 'SwissProt only article collection' include articles containing interaction pairs that can be normalized to SwissProt entries. The table shows the comparative results for the extraction of interaction pairs.