Classification of large biomolecular structures by mapping of shape and deformation parameters

The classification of biomolecule structures is a longstanding issue, mainly motivated by the need for efficient similarity criteria in connection to the study of their biological activity and the folding process. The labelling of structures with appropriate invariant parameters, uniquely associated to a given molecular geometry, can induce structure classification, leading to the emergence of patterns and regularities. Statistics of such parameters can help in testing the accuracy of the experimentally measured and theoretically predicted structures. The recent successful use of computational approaches based on machine learning and neu-ral networks further motivates efforts in such direction. We consider here to approach the classification of protein structures by assigning to them, in a unique way, sets of invariant quantities. These are shape parameters and deformation indexes derived from “symmetric” hyperspherical coordinates, as introduced by us in previous works.