Benchmarking of protein descriptor sets in proteochemometric modeling (part 2): modeling performance of 13 amino acid descriptor sets

Table 2 Data sets used for the bioactivity benchmarks

	ACE inhibitors	GPCRs	NNRTIs	PIs
Total size (data points)	58	6,046	4,024	6,995
Total compounds	n/a	3,230	451	9
Average compound tanimoto distance (ECFP_6)	n/a	0.92	0.54	0.73
Average euclidian distance compounds (physicochemical)	n/a	1.28	n/a	0.90
Total targets (peptides / proteins)	58	32	14	1060
Average target tanimoto distance (ProtFP (Feature))	0.83	0.22	0.14	0.03
Average euclidian distance target (ProtFP (PCA3))	1.35	0.93	0.44	0.26
Completeness (% of total compound - target pairs)	n/a	0.06	0.64	0.73

The datasets were selected to obtain a diverse collection of sets amenable to PCM modeling. To this extend both GPCRs and enzymes were included. Moreover sets representing initial hit discovery, lead optimization, and well-established structure-activity space modeling were included.

ISSN: 1758-2946