Universal architectural concepts underlying protein folding patterns

Arthur M. Lesk, Ramanan Subramanian, Lloyd Allison, David Abramson, Peter J. Stuckey, María García de la Banda, Arun S. KonagurthubioRxiv 2018 · DOI

Abstract

What is the architectural ‘basis set’ of the observed universe of protein structures? Using information-theoretic inference, we answer this question with a comprehensive dictionary of 1,493 substructural concepts. Each concept represents a topologically-conserved assembly of helices and strands that make contact. Any protein structure can be dissected into instances of concepts from this dictionary. We dissected the world-wide protein data bank and completely inventoried all concept instances. This yields an unprecedented source of biological insights. These include: correlations between concepts and catalytic activities or binding sites, useful for rational drug design; local amino-acid sequence–structure correlations, useful for ab initio structure prediction methods; and information supporting the recognition and exploration of evolutionary relationships, useful for structural studies. An interactive site, PROÇODIC, provides access to and navigation of the entire dictionary of concepts.

Introduction

Proteins are the workhorses of life, and their function is dictated by their complex three-dimensional shapes. While we understand the basic levels of protein structure—the linear sequence of amino acids (primary) and the local helices and strands (secondary)—the "middle ground" of how these elements assemble into recurrent 3D patterns (supersecondary and tertiary) has remained elusive. Previous attempts to categorize these intermediate building blocks were often limited by human bias or focused on very short, local fragments. There was no statistically rigorous, objective "dictionary" that could explain the entire known universe of protein folds. This research sought to find the fundamental "basis set" of protein architecture: the smallest set of structural concepts that can most efficiently describe all known protein structures.

Methodology

The researchers treated the discovery of protein building blocks as a lossless data compression problem. Using the Minimum Message Length (MML) criterion, they argued that the "best" dictionary of structural concepts is the one that allows for the shortest possible description (maximum compression) of the entire protein database.

Protein folds were represented using "tableaux"—concise symmetric matrices that capture the order, geometry, and contact information of secondary structural elements (SSEs) along the amino acid chain. An unsupervised machine learning approach using Simulated Annealing was developed to search the astronomical space of possible dictionaries. The goal was to find a set of concepts that balances complexity (the size of the dictionary) with fidelity (how well the concepts explain observed structures).

Protein dissection into structural concepts

As illustrated in Figure 1, any protein structure (such as 1ahq) can be dissected into instances of these concepts. The initial dictionary was inferred from the ASTRAL SCOP-95 dataset (26,949 domains) and subsequently used to inventory every instance of these concepts across the entire world-wide Protein Data Bank (wwPDB), which contained over 140,000 entries at the time of the study.

Key Results and Insights

The final dictionary, termed PROÇODIC, contains 1,493 unique concepts, ranging in size from 2 to 28 secondary structural elements. These concepts account for 66.35% of all amino acids in the dissected wwPDB. The remaining ~33% consists of "null" regions—structurally variable loops or rare arrangements that do not fit into the topologically-conserved definition of a universal concept.

Cumulative amino acid coverage of the concept dictionary

Figure 5 shows the cumulative coverage of the dictionary. This high level of coverage suggests that the 1,493 concepts indeed form a robust "basis set" for known protein architecture. Beyond mere classification, the study yielded several critical biological insights:

  • Function Follows Form: Many concepts are "hard-wired" to specific biological roles. The authors identified 463 concepts highly correlated with specific ligand-binding sites or catalytic activities. For example, concept c_0060 is intrinsically linked to the proteolytic active sites of proteasomes.
  • Sequence-Structure Correlation: Nearly 80% of these structural concepts have strong, detectable amino-acid sequence preferences. This suggests that the 3D "concept" is often encoded directly into the local sequence, a vital insight for ab initio structure prediction methods.
  • Predictive Potential: In tests on CASP12 targets, the dictionary supported structure prediction effectively; for 26 out of 33 proteins, more than 50% of the target sequence had detectable similarity to known concept usages.

Conclusion and Future Work

PROÇODIC provides a comprehensive, objective library of the building blocks of protein architecture. By mapping these structural motifs to specific biological functions and sequence patterns, it provides a powerful new framework for drug design and protein engineering. Future work will focus on addressing the "null" space of structurally variable regions and refining automated ab initio prediction tools based on these universal concepts. The entire dictionary and dissection tools are available via the interactive PROÇODIC web resource.

Summary updated

Abstract

ABSTRACT What is the architectural ‘basis set’ of the observed universe of protein structures? Using information-theoretic inference, we answer this question with a comprehensive dictionary of 1,493 substructural concepts . Each concept represents a topologically-conserved assembly of helices and strands that make contact. Any protein structure can be dissected into instances of concepts from this dictionary. We dissected the world-wide protein data bank and completely inventoried all concept instances. This yields an unprecedented source of biological insights. These include: correlations between concepts and catalytic activities or binding sites, useful for rational drug design; local amino-acid sequence–structure correlations, useful for ab initio structure prediction methods; and information supporting the recognition and exploration of evolutionary relationships, useful for structural studies. An interactive site, P roçodic , at http://lcb.infotech.monash.edu.au/prosodic (click) provides access to and navigation of the entire dictionary of concepts, and all associated information.
BibTeX
@misc{lesk2018universal,
  author = {Arthur M. Lesk and Ramanan Subramanian and Lloyd Allison and David Abramson and Peter J. Stuckey and María García de la Banda and Arun S. Konagurthu},
  title = {{Universal architectural concepts underlying protein folding patterns}},
  year = {2018},
  doi = {10.1101/480194},
  howpublished = {bioRxiv},
}
Figure