The ability of a protein to perform its function is mainly dened by the spatial shape it exists in and the way the protein alternates between several stable shapes. To prevent or cure diseases related to protein malfunctioning we study the conformational space of proteins. Sampling-based motion planning algorithms from the eld of robotics have been very successful at this task. However, studying the conformational space of large proteins with hundreds or thousands of Degrees of Freedom remains a big challenge. In this work we investigate how the dimensionality curse can be mitigated by means of low-dimensional projections. Our experiments demonstrate that incorporating the information available on the studied protein into the projection can benefit the conformational exploration process. The techniques we developed to generate efficient low-dimensional projections can enable sampling-based planners to study protein systems, such as viruses, that are currently too large to be investigated by other methods.