Hierarchical Ensemble Methods for Protein Function Prediction

Table 2

Comparison of the results (average per class 𝐹-scores) achieved with single sources and multisource (data fusion) techniques. FLAT, HTD, HTD-CS, HB (HBAYES), HB-CS (HBAYES-CS), TPR, and TPR-W ensemble methods are compared with and without data integration. In the last row, the number in parentheses refers to the percentage relative increment in 𝐹-score performance achieved with data fusion techniques with respect to the best single source of evidence (BIOGRID).

Methods	FLAT	HTD	HTD-CS	HB	HB-CS	TPR	TPR-W
Single-source
BIOGRID	0.2643	0.3759	0.4160	0.3385	0.4183	0.3902	0.4367
String	0.2203	0.2677	0.3135	0.2138	0.3007	0.2801	0.3048
PFAM BINARY	0.1756	0.2003	0.2482	0.1468	0.2407	0.2532	0.2738
PFAM LOGE	0.2044	0.1567	0.2541	0.0997	0.2847	0.3005	0.3160
Expr.	0.1884	0.2506	0.2889	0.2006	0.2781	0.2723	0.3053
Seq. sim.	0.1870	0.2532	0.2899	0.2017	0.2825	0.2742	0.3088
Multisource (data fusion)
Kernel fusion	0.3220(22)	0.5401(44)	0.5492(32)	0.5181(53)	0.5505(32)	0.5034(29)	0.5592(28)