Exploiting large-scale drug-protein interaction information for computational drug repurposing

BMC Bioinformatics

Table 1 Impact of SEA-predicted drug-protein interactions on Type II model performance ^a

		With SEA		Without SEA
Evaluation of top-ranking scores (%)	Drugs in training set	Fraction	σ	Fraction	σ
High Blood Pressure model
1	1	0.14	0.06	0.15	0.07
5	1	0.36	0.11	0.36	0.11
10	1	0.50	0.14	0.48	0.14
1	2	0.14	0.06	0.18	0.07
5	2	0.48	0.09	0.45	0.10
10	2	0.64	0.09	0.60	0.11
1	3	0.19	0.06	0.20	0.07
5	3	0.53	0.08	0.52	0.10
10	3	0.70	0.08	0.67	0.10
HIV model
1	1	0.16	0.09	0.20	0.19
5	1	0.28	0.14	0.30	0.18
10	1	0.37	0.17	0.39	0.21
1	2	0.17	0.07	0.18	0.07
5	2	0.32	0.12	0.33	0.10
10	2	0.44	0.13	0.43	0.15
1	3	0.19	0.07	0.21	0.07
5	3	0.34	0.11	0.37	0.09
10	3	0.48	0.10	0.50	0.12
Malaria model
1	1	0.14	0.07	0.20	0.24
5	1	0.34	0.14	0.54	0.25
10	1	0.70	0.22	0.79	0.22
1	2	0.20	0.13	0.20	0.13
5	2	0.43	0.20	0.58	0.16
10	2	0.73	0.15	0.89	0.10
1	3	0.21	0.09	0.21	0.10
5	3	0.49	0.15	0.66	0.13
10	3	0.77	0.13	0.91	0.07

^aThe type II models were built with one, two, and three positive drugs in the positive class of the training set. The fraction of positive drugs in the baseline class that scored in the highest 1%, 5%, and 10% of the compounds are recorded for models using the STITCH 3.1 database with and without SEA-predicted drug-protein interactions. Fraction, fraction of known drugs retrieved; σ, standard deviation; SEA, similarity ensemble approach.

ISSN: 1471-2105