Pattern Recognition and data mining on protein surfaces

Many methods have been developed to analyze protein primary sequences or overall tertiary structures since sequences and/or structures similarity may imply common functional characteristics or active properties. However, evolution can lead sequences to diverge or structures to change topology even though retaining surface determinants essential to protein function. In such cases sequence or structure comparisons could not be sufficient to identify functions and evolutionary relationships among proteins. On the other hand, protein surface comparisons could identify those functions determinants that are independent of sequence or 3D structure. Here, we propose an original approach for comparison of proteins based on pattern recognition on surfaces, which has the advantage of being unsupervised and not based on already known functional meanings. Our method initially determines homogeneous patches on the protein surfaces, considering electric potential, hydrophobicity and shape index as distinguishing features. Then, complex patterns, composed by sets of patches that assume a similar spatial arrangement in different proteins, are detected. The method has been tested on a benchmarck database of 98 protein domains. The database has been constructed starting from 4 reference protein domains that have been subjected to progressive surface aminoacids mutations. The domains have been correctly classified and the progressive distance beetwen proteins, due to mutations, has been emphasized.