The "Druggability" article on Wikipedia (https://en.wikipedia.org/wiki/Druggability) provides multiple references to druggability assessment techniques. They all somehow quantify druggability from the properties of cavities detected on the protein surface. None of them can really be deemed "reliable" though. Indeed, saying that a given pocket is undruggable is equivalent to saying that any ligand would have low affinity to it. Accurate prediction of protein-ligand affinity (for a single protein-ligand pair) is a matter of weeks, while the "druggability" assessment is usually a matter of seconds/minutes (sounds like a free lunch). I would rather rely on simple rules of thumb. Does the protein of interest has any structure (i.e., is not disordered)? Does it bind any endogenous small-molecule (e.g., an enzyme co-factor) or a peptide?.If all answers are "yes" then there is a good chance (though not 100%) that the protein is druggable, whatever the druggability index says. Inversely, id the answers are "no" the chances of druggability are likely to be close to zero (though not zero), whatever the druggability index says.
Schrödinger is promoting WaterMap as an approach to evaluating the druggability of a target, where enzymes with continuous stretches of entropic waters are deemed druggable. There are alternatives to watermap as well from other vendors, such as SZmap, if your institution does not have the Schrödinger licence
3decision is another solution worth mentioning. It identifies all pockets on your protein structure and gives you a druggability score for each of them.
Once you've found a drubbale pocket, 3decision can help you retrieve all relevant structural knowledge about your target, e.g. known mutations, similar homologues, potential off-targets etc. I use this in our target assessment reports.