Development of machine learning classifiers to predict compound activity on prostate cancer cell lines

Table 4 Prediction results on isolated ChEMBL assays

Assay ChEMBL ID	N compounds	N active	N inactive	Accuracy	TN	FP	FN	TP	TNR	FPR	FNR	TPR
PC3
CHEMBL4155048 [43]	33	33	0	0.94	0	0	2	31	–	–	0.06	0.94
CHEMBL1840268 [44]	34	28	6	0.71	1	5	5	23	0.17	0.83	0.18	0.82
CHEMBL3097567 [45]	38	0	38	0.63	24	14	0	0	0.63	0.37	–	–
CHEMBL2040207 [46]	44	31	13	0.41	9	4	22	9	0.69	0.31	0.71	0.29
CHEMBL3062059 [47]	47	24	23	0.77	20	3	8	16	0.87	0.13	0.33	0.67
CHEMBL3788954 [48]	64	23	41	0.63	33	8	16	7	0.80	0.20	0.70	0.30
DU145
CHEMBL4686002 [49]	30	0	30	0.90	27	3	0	0	0.90	0.10	–	–
CHEMBL4158478 [50]	38	26	12	0.69	6	6	6	20	0.50	0.50	0.23	0.77
CHEMBL2406593 [51]	50	0	50	0.28	14	36	0	0	0.28	0.72	–	–

RF classifiers (GAP 20, PCC = 95) results on the external test set consisting of the compounds from a single ChEMBL assay for PC3 and DU145. Reference articles of each assay are added in brackets

ISSN: 1758-2946