U-Net is an encoder-decoder neural-network architecture with skip connections. Its encoder downsamples inputs into hierarchical features, while its decoder upsamples and combines information across spatial scales. section 11.3.1
For protein contact prediction, U-Net converts multiple-sequence-alignment features into a contact-probability matrix. Its multiscale design captures local patterns from secondary structures and longer-range patterns from domains. section 11.3.1
A contact-prediction workflow uses U-Net as follows:
- Build a multiple sequence alignment from homologous proteins. section 11.3.1
- Provide 72 alignment-derived features, including amino-acid frequencies, mutual information, and direct-coupling statistics. section 11.3.1
- Convert predicted contact probabilities into graph edges, commonly using a threshold of 0.5. section 11.3.1
Figure 11.5 Transforming a Protein's 3D Structure into a 2D Graph. Converting a protein's spatial information into a graph representation. (a) A simplified 3D protein structure with dashed lines connecting amino acid residues that are in close spatial proximity (a 'contact'). (b) Spatial information is translated into a contact map, which is a 2D matrix. A point at (i, j) is marked if residue i and residue j are in contact, serving as an adjacency matrix for the graph. (c) The contact map is used to construct a graph where the amino acid residues are the nodes and the contacts are the edges. This graph captures the key residue interactions in a format that is invariant to the protein's rotation and translation in 3D space.

Trace how spatial residue contacts become a two-dimensional contact map and then a protein graph. section 11.3.1






