Clustering techniques for protein surfaces

Though most approaches to protein comparison are based on their structure, several studies produced evidence of a strict correlation between the surface characteristics of proteins and the way they interact. Surface-based techniques for protein comparison typically require applying clustering algorithms to the punctual 3D description of the surface in order to produce a compact surface representation, capable of effectively condensing its description. In this paper, we propose a formalization of the requirements for surface clustering in the biochemical context and present two different clustering techniques that meet them, based, respectively, on region-growing and on an original template matching algorithm. We discuss the validity of these techniques with the support of tests performed on a set of about one hundred protein models generated by punctual mutations of four structurally characterized proteins. Finally, an analysis is made of how different factors impact on the effectiveness of clustering in capturing surface similarities.