Topic automatically creating for discussing designs at:
https://covid.postera.ai/covid/submissions/672ec752-1201-422d-83dc-d40faf5ded3c
I took these structures and docked them in the catalytic dyad with Autodock Vina. The plot below shows these structures as impulses, against the background distribution from docking the 100’000 GB-GA combined fragments. As you can see, they are all better than the median from docking. I thought this was interesting, as it shows that very diverse approaches (graph-convolution-network, on the fragment data including zero-scores, and docking) provide a fairly identical view.
(Nb: unfortunately some of Jarryl’s structures score identically, and their presence is obscured in the graph as the impulses sit on top of one another.)