Matched pairs demonstrate robustness against inter-assay variability

Table 2 Comparison of matched pair data with the original manuscript results [1]

Curation level	Type	MAE	F > 0.3	F > 1.0	Assay pairs	Compound pairs
IC₅₀ minimal	Ref	0.50	0.64	0.27	1,358	38,022
IC₅₀ minimal	Pair	0.33	0.54	0.12	1,338	372,683
IC₅₀ maximal	Ref	0.27	0.48	0.13	26	340
IC₅₀ maximal	Pair	0.18	0.34	0.06	24	1,773
K_i minimal	Ref	0.52	0.67	0.30	587	7,734
K_i minimal	Pair	0.36	0.56	0.18	584	36,189
K_i maximal	Ref	0.47	0.69	0.32	282	2,434
K_i maximal	Pair	0.40	0.62	0.43	279	2,900
K_i maximal pruned	Ref	0.45	0.58	0.25	9	115
K_i maximal pruned	Pair	0.03	0.21	0.08	8	311

Note that the compound pairs for Pair-type datasets refers to cross-assay matched structural pairs, rather than pairs of individual compounds as in the Ref-type datasets. Additionally, reference pruning was less strict, leaving some Carbonic Anhydrase assays. These were removed in our workflow after identifying a cluster of outliers (Figure S4)

ISSN: 1758-2946