Assessing 3D scores for protein structure fragment mining

Frédéric Guyon1, Pierre Tufféry1,21MTi, INSERM UMR-S973, Université Paris Diderot-Paris 7, Paris, France; 2RPBS, Université Paris Diderot-Paris 7, Paris, FranceAbstract: Quantifying the 3D similarity between two proteins is a difficult task that has motivated the assessment of several 3D scores. New developments in protein modeling and analysis have led to the emergence of new interest towards mining structures at the local level. We assess in the context of fragment mining several dissimilarity scores. We revisit the concept of mirror conformation previously introduced at the level of complete structures and extend it to the more local level. We also consider an explicit criterion measuring the fragment boundary discrepancies. Whereas classical criteria such as the root mean square deviation (RMSd) fail to identify similar shapes in a consistent way, we show that local mirror and boundary mismatch filtering greatly supplements classical scores to select significant matches. The geometrical conditions defined by such criteria can be considered as signatures of fragment similarity. Furthermore, it is possible to tune the degree of similarity depending on the size of the mirrors accepted. This results in a more intuitive perception of the concept of similarity, and opens new perspectives for the rapid mining of large collections of structures.Keywords: protein fragments, similarity, distance, mining