Recognition and association plays a crucial role in the assembly and function of transmembrane proteins. Despite significant research in the last years, these processes are only partially understood. We investigate the association of model transmembrane alpha-helices embedded in lipid bilayers by means of uniform sampling along their separation distance and accurate potential of mean force (PMF) calculations. The model system consists of a pair of alpha-helices of Glycophorin A, a system that experimentally exhibits dimerization in a lipid membrane. Using established coarse-grained models, we develop an efficient Monte Carlo methodology to overcome sampling limitations imposed by long characteristic times present in conventional lipid membrane simulations. Designed Monte Carlo moves aim to increase conformational sampling for the helices in the hydrophobic environment. A combination of the Expanded Ensemble Density of States formalism and hybrid molecular dynamics allow for efficient and accurate calculations on the association of the helices. Using our methodology, we evaluate the effect of different lipid composition on the mechanism of the recognition process of protein molecules in a hydrophobic environment.