Drug-spaces of nine crystallographic protein / ligand models have been comparatively explored by including Toxicity Risk assessment during computational co-evolution. Tens of thousands children were randomly generated from parent ligands and iteratively selected for higher affinities, increased specificities and low Toxicity Risk using DataWarrior / Build Evolutionary Library algorithms, mimicking natural evolution. Only a few hours of co-evolution increased ~ 2-fold the numbers of non-toxic children. Top-leads predicted drug-like properties, nanoMolar affinities (confirmed by AutoDockVina), higher specificities, absence of known toxicities, and similar docking to their initial binding cavities. Tables were provided with multi-threshold-adjustable filters for alternative in silico explorations of this new "co-evolutionary docking" tool.