Four nanoseconds can be sufficient, providing that (i) you run a set of such MD simulations starting from different initial conditions, e.g., by varying the seed for generating the initial velocities, and (ii) the calculated protein-ligand binding free energy shows low uncertainty (below 2 kcal/mol).

See how long the unrestrained ligand stays in the binding site in each MD simulation in the set of (let's say) 10. If the complex is stable for 4 ns on average, you will have 40 ns, and if you collect non-bonded interaction energies every 20 fs, you will have an ensemble of 2 million energies -- and that might be sufficient. If the uncertainty in the calculated protein-ligand binding free energy is too high: (a) increase the size of the set, or (b) apply a weak positional or distance restraint.

You did not announce the goal of your simulation study. It is difficult to help without knowing it.