Reconstruction of Protein Structures from a Vectorial Representation

We show that the contact map of the native structure of globular proteins can be reconstructed starting from the sole knowledge of the contact map’s principal eigenvector, and present an exact algorithm for this purpose. Our algorithm yields a unique contact map for all $221$ globular structures of PDBselect25 of length $N\le 120$. We also show that the reconstructed contact maps allow in turn for the accurate reconstruction of the three-dimensional structure. These results indicate that the reduced vectorial representation provided by the principal eigenvector of the contact map is equivalent to the protein structure itself. This representation is expected to provide a useful tool in bioinformatics algorithms for protein structure comparison and alignment, as well as a promising intermediate step towards protein structure prediction.

Figure 1

The protein amylase/serine protease inhibitor with PDB id. 1bea. Shown are the following: (a) The three-dimensional structure (drawn using MolScript and Raster3D), (b) the CM $\mathbf{C}$ (black means $C_{ij}=1$ and white means $C_{ij}=0$), and (c) the PE $\mathbf{v}$. The reduction from (a) to (b) is done by a distance threshold, and the one from (b) to (c) is by a diagonalization. The reconstruction from (c) to (b) is described in this Letter, whereas the one from (b) to (a) is described in Ref. 8.