In drug discovery, it is essential to identify binding sites on protein surfaces that drug-like molecules could exploit to exert a biological effect. Both X-ray crystallography and NMR experiments have demonstrated that organic solvents bind precisely at these locations. We show that this effect is reproduced using molecular dynamics with a binary solvent. Furthermore, analysis of the simulations give direct access to interaction free energies between the protein and small organic molecules, which can be used to detect binding sites and to predict the maximal affinity that a drug-like molecule could attain for them. On a set of pharmacologically relevant proteins, we obtain good predictions for druggable sites as well as for protein-protein and low affinity binding sites. This is the first druggability index not based on surface descriptors and, being independent of a training set, is particularly indicated to study unconventional targets such as protein-protein interactions or allosteric binding sites.