The invention provides computer-implemented methods and apparatus implementing a hierarchical protocol using multiscale molecular dynamics and molecular modeling methods to predict the presence of transmembrane regions in proteins, such as G-Protein Coupled Receptors (GPCR), and protein structural models generated according to the protocol. The protocol features a coarse grain sampling method, such as hydrophobicity analysis, to provide a fast and accurate procedure for predicting transmembrane regions. Methods and apparatus of the invention are useful to screen protein or polynucleotide databases for encoded proteins with transmembrane regions, such as GPCRs.