Generation of Gene Ontology benchmark datasets with various types of positive signal

BMC Bioinformatics

Table 3 Evaluation of DAVID and GENERATOR clustering using the generated test datasets

dataset num.	PPV GEN.	PPV DAVID	Sens. GEN.	Sens. DAVID	F1 GEN.	F1 DAVID	GEN. -DAVID
1	0.3	0.09	0.6	1	0.4	0.17	0.23
2	0.38	0.06	0.75	1	0.5	0.12	0.38
3	0.38	0.07	0.6	0.8	0.46	0.13	0.33
4	0.36	0.04	0.8	0.4	0.5	0.07	0.43
5	0.33	0.07	0.6	1	0.43	0.14	0.29
6	0.23	0.05	1	1	0.38	0.1	0.28
7	0.14	0.04	0.33	0.67	0.2	0.07	0.13
8	0.15	0.02	0.67	0.33	0.25	0.04	0.21
9	0.36	0.03	0.8	0.4	0.5	0.05	0.45
10	0.38	0.09	0.6	1	0.46	0.16	0.3
11	0.67	0.07	0.8	0.8	0.73	0.13	0.59
12	0.42	0.05	1	0.6	0.59	0.09	0.5
13	0.43	0.07	0.6	0.8	0.5	0.13	0.38
14	0.5	0.04	1	0.5	0.67	0.07	0.6
15	0.5	0.09	1	1	0.67	0.16	0.51
16	0.33	0.03	0.67	0.67	0.44	0.06	0.38
17	0.09	0.03	0.33	0.67	0.14	0.07	0.08
18	0.42	0.07	1	1	0.59	0.13	0.46
19	0.38	0.05	0.75	0.75	0.5	0.09	0.41
20	0.13	0.03	0.33	0.67	0.18	0.06	0.12

Table presents the Positive Predictive Value (PPV), sensitivity (sens.) and F1 score for GENERATOR(GEN.) and DAVID clustering results. Furthermore, we show the difference of F1 scores between two methods. Notice that the F1 score for GENERATOR is consistently better, as is shown by the difference in the last column. DAVID outperforms GENERATOR only in sensitivity, but this is natural as DAVID created 3-10 times more clusters.

ISSN: 1471-2105