Fig. 3: Validation of specific compound-dual-target activities using POLYGON.
From: De novo generation of multi-target compounds using deep generative chemistry
a A library of 109,811 compounds was scored for activity against pairs of protein targets. (Compound, Target1, Target2) triplets for which low IC50 was observed against both targets (both activities <1 µM) were classified as “dual activity observed”, otherwise as “dual activity not observed”. Similarly, triplets for which low IC50 was predicted against both targets were classified as “dual activity predicted”, otherwise as “dual activity not predicted”. Observed and predicted dual activities were compared to tabulate true positive, false positive, false negative, and true negative cases. b Contingency table for classification of dual activity status of (Compound, Target1, Target2) triplets. Odds ratio computed by two-sided Fisher Exact Test and p-value computed by chi-squared test; exact p-values are provided in the Source Data file. c Precision-recall of POLYGON classification of dual active compounds at various IC50 activity thresholds (colored dots). d Receiver-operator characteristic of POLYGON classification of dual active compounds at various activity thresholds (colored dots). Source data are provided as a Source Data file.
