liveBook · Manning

Protein graph

As explained in Build AI Drug Discovery Pipelines

A protein graph is a graph-structured representation of a protein in which biologically relevant protein entities and relationships are encoded as nodes and edges.

A protein graph represents amino acid residues as nodes and spatial contacts as edges. A contact map supplies its adjacency matrix: entry C[i,j] is 1 when residues are sufficiently close. section 11.3.1

Construct the representation by:

  • Computing an L×L contact map for a protein sequence of length L. section 11.3.1
  • Defining contact with an 8 Å distance threshold between C-beta atoms, or C-alpha atoms for glycine. section 11.3.1
  • Converting qualifying matrix entries into graph edges and adding sequence-derived residue features. section 11.3.1

From structure or sequence

A Python workflow loads a .npy contact map, checks that its dimensions match the target sequence, extracts edges using a configured threshold, and computes features from an alignment file. When structures are unavailable, deep-learning contact predictors can estimate maps from sequence information. section 11.3.1

How Build AI Drug Discovery Pipelines covers it

How deep it goes
Explains contact-map construction, invariance, sequence prediction, and Python edge extraction, but defers full feature computation details.
Examples
Uses an 8 Å contact rule, an L×L matrix, and Python files containing contact maps and alignments.
What it assumes
Uses protein sequences, contact maps, adjacency matrices, and alignments while defining their roles operationally.
What you can do after
Builds a protein graph from a contact map and sequence-derived features using a Python workflow.
How it connects
Connects contact maps, adjacency matrices, amino acid residues, multiple sequence alignments, and deep learning.

Related concepts

This explainer was generated by Manning's knowledge graph from the text of Build AI Drug Discovery Pipelines. The section links point to the parts of the book it draws on.