Table 4.

Average enrichment in the simulated target-prediction experiment when training with random negatives

Enrichment (%)	PS-IRV	SVM	RF
$1\mathrm{\mu }\text{M}$ cutoff
5	96	92	95
10	98	94	96
20	98	97	97
30	99	98	97
$5\mathrm{\mu }\text{M}$ cutoff
5	95	92	95
10	97	94	96
20	98	97	97
30	98	98	97
$10\mathrm{\mu }\text{M}$ cutoff
5	94	94	93
10	96	95	96
20	97	97	97
30	97	98	97

Models are tested using 10-fold cross-validation. 9000 randomly selected ChEMBL molecules are added to the original test set as putative inactives. 1000 randomly selected ChEMBL molecules are added to the original training sets as putative inactives. Best results at each cutoff are in italics