G-protein coupled receptors (GPCRs) are allosteric membrane proteins mediating cellular signaling. GPCRs exhibit multiple inactive and active conformations, and the population balance between these conformations is altered upon binding of signaling molecules (or ligands). However, the nature of the conformational ensemble or the mechanism of the conformational transitions is not well understood. We present a multiscale computational approach combining a coarse-grained discrete conformational sampling method with fine-grained molecular dynamics investigating the effect of various ligands binding on the ensemble of conformations sampled by human beta 2-adrenergic receptor (beta 2AR). We show that the receptor, in the absence of any ligand, samples an extensive conformational space that includes breathing of the orthosteric ligand binding site and shear motion of the transmembrane helices 5 and 6 against the other helices. The shear motion is similar to the reorganization of the intracellular regions of TM3, TM5, and TM6 observed in the crystal structure of the active state of GPCRs. The binding of agonist norepinephrine or partial agonist salbutamol leads to the selection of a subset of conformations including active and inactive state conformations, while inverse agonist carazolol selects only inactive state conformations. The dynamics of water observed during the simulations provides an explanation for the conformational changes observed in the solution-based fluorescence spectroscopic measurements on agonist activated beta 2AR, which could not be explained by the agonist bound beta 2AR crystal structure. This study shows that the receptor activation depends on both the low energy states and the range of the conformations sampled by the receptor.