Most existing clustering approaches require the complete graph information, which is often impractical for large-scale protein-protein interaction networks. We proposed a novel algorithm which does not embrace the universal approach but instead tries to focus on local ties and model multi-scales of biological interactions in these networks. It identifies functional leaders and modules around these leaders using local information. It naturally supports overlapping information by associating each node with a membership vector that describes its involvement of each community. In addition to uncover overlapping communities, we can describe different multi-scale partitions allowing to tune the characteristic size of biologically meaningful modules. The high efficiency and accuracy of the proposed algorithm make it feasible to be used for accurately detecting community structure in real biomolecular networks.