Protein secondary structure classification for PPI site prediction using ASPRA distance

Proteins perform a broad range of functions within organisms. Most of these activities are possible by interacting with other molecules, including proteins, RNAs or DNAs, and small ligands. The interactions between two proteins referred to as protein-protein interactions (PPIs) determine the metabolic and signaling pathways, whose dysfunction or malfunction cause several diseases, among which neurodegenerative disorders and cancer. Proteins consist of one or more sequences of amino acids that fold back on themselves by determining three-dimensional configurations or shapes. Such a configuration is strictly related to its function because it determines whether the protein can interact with other molecules. A shape can be abstracted as secondary structure that are widely employed in structural biology applications since they can be handled efficiently and are relevant from a biological perspective. They are also used for comparison and classification. In this work, we propose to use the Algebraic Structural Pseudoknot RNA Alignment (ASPRA) distance to quantify the structural differences among proteins and obtain new protein structural classifications by classical machine learning techniques in the scenario of supporting an incrementing of the performance of PPI site prediction techniques structural based motivated by the analysis of our recent results. It is evident that the performance of our PPI site prediction approach depends on the similarity of protein architecture: it is significant when the proteins of dataset show similar architecture. We represent the secondary structures as an arc annotated sequence, and split each residue into as many residues as the number of bonds it forms without adding new crossings among arcs since ASPRA distance requires that each amino acid is involved in at most one base pair. This requirement means that only one arc can be attached to each vertex. We have developed the proposed approach in Python by using Biopython and DSSP packages. Our approach allows classify allows us to predict PPI site interaction of proteins having similar structures and organize appropriately the protein dataset.