macromol-census 0.3.0

Tools for creating machine-learning datasets from macromolecular structure

Macromolecule Census is a tool for identifying high-quality, non-redundant subsets of the biological assemblies in the protein data bank (PDB). A particular emphasis is to accommodate all kinds of macromolecules; not just proteins. Briefly, this process involves the following steps:

• Rank each structure by metrics including clash score, resolution, $R_{free}$, Q-score, NMR restraints.

• Rank each assembly by biological relevance, subchain cover, and size.

• Cluster protein, DNA, and RNA molecules by sequence similarity.

• Cluster small molecules and branched polysaccharides by identity.

The primary use-case for this software is the creation of datasets for machine learning. This typically entails iterating through each assembly in ranked order, adding unique training examples as they appear.