Derek has a good post that takes a philosophical approach toward the whole question of "druggability". The main question is; given a disease state and biochemical knowledge of all the mechanisms involved in that state, can you find a therapeutic that will abolish the disease state and restore another one which we ordinarily term "healthy"? The whole thing is worth reading.
For the moment let's make things simple and assume that the therapeutic we are looking for is a small molecule. Derek's post actually takes me back to my post about why physics cannot solve the problem of drug design. The main challenge in drug discovery is that we have to come up with solutions that have to extend over a whole range of emergent phenomena; an ideal drug will have to modulate every level of biological organization from the whole body down to organs systems, cells and molecular targets. Some of our best drugs do this but that was a happy and accidental coincidence. We still cannot deliberately design in features that will modulate a system across multiple emergent levels.
However I am hard pressed to see why this cannot be done in principle. That is because in some sense, the problem of druggability comes down to the simpler problem of "ligandability"; that is, for every given protein and every arbitrary binding site, can you find a complementary key that fits the lock? My guess is that the answer to this question is a yes since ultimately it boils down to forces and geometric complementarity. Can we design a small molecule that fills a pocket, forms hydrogen bonds and electrostatic interactions and gains binding affinity through the hydrophobic effect by displacing water molecules? I would tend to think that the answer is a yes across the board, so I don't see why the problem shouldn't be solvable for every single protein at least in principle.
There's other challenges in drug development of course; maintaining the right blood levels, modulating half lives and protein-drug binding kinetics, avoiding drug-drug interactions, but all of these problems have at their root the interaction between molecules which can be modulated. Ultimately, every drug works its magic through molecular interactions, even if they are spread across multiple emergent levels and targets. The problem of druggability is in one sense the problem of ligandability stated multiple times. Solving the problem of druggability is tantamount to solving the problem of ligandability for several targets; the target of interest, anti-targets like hERg and cytochrome P450, a judiciously picked subset of targets whose simultaneous inhibition leads to the required beneficial effect (like the targets for some of the "selectively non-selective" kinase inhibitors out on the market). Want to improve off-rates? Improve protein-ligand interactions. Want to maintain drug levels? Minimize off-target degradation by proteases or esterases. Want to avoid interactions with other drugs? Regulate or improve inhibition of cytochrome P450s or PgP. If the problem of ligandability can be solved for multiple targets, wouldn't it be equivalent to solving the problem of druggability?
In practice of course things are very different since it may well be impossible to practically satisfy every single constraint leading to the abolition of a specific disease state. Since biological systems are emergent we could also get very unexpected feedback from this kind of perturbation that throws our predictions off. I also tend to agree with Derek that we can look through every SMILES string that we have and still not find the right molecule for addressing the multiple-ligandability problem. We can throw every single molecular library in the world at Ras and still not find anything worthwhile. But knowing what we do about the physical and chemical principles that govern biological systems, I doubt it would be so because we are dealing with the biological version of Hilbert's halting problem which Turing proved was undecidable. It could still be because we have simply not tried looking everywhere.