Authors have carried out new designs of molecules that could bind to EVOB (Ebola virus). They used the known binder BCX4430 and then generated molecules using deep learning package LigDream. Followed by this, authors have performed docking and identified two molecules with higher docking score than BCX4430 and thereafter performed various analysis including MD simulations and MM/PBSA calculations.
While the approach is straightforward, I have some queries:
(a) Have authors performed MD simulations on some of the poorly docked molecules to see that those are indeed not remaining bound to the active site
(b) The MM/PBSA calculations seems to be not converged even after 100 ns for mol1. The error bar for 50ns simulation is much smaller. Yet, the values change for 100ns simulation. This cast a strong doubt on the free energy values reported. Authors need to show a plot of binding energy with respect to time to show the convergence.

(c) If authors used mol1_069 and used LigDream, would they get some more strong binders? This is worth a try.

Authors have not understood my first comment. The primary assessment of the binding of the generated ligand was done here by docking. However, docking has several drawbacks -- such as absence of water, flexibility of the receptor, etc. Therefore, as authors have shown the free energy of its strongest binders, they should also show free energy of their weaker binders also to show that docking score is somewhat consistent with the free energy.