The Nature of Protein Hierarchies and Implications for Hierarchical Modelling: An AI-Assisted Computational Study with Statistical Mechanics, Information Theory and Unsupervised Learning

Proteins are widely described as hierarchies of nested structural units, and hierarchical models are widely used to simulate them, yet what kind of hierarchy proteins actually possess, and what that hierarchy implies for how they should be modelled, has not been established. This work addresses these two questions directly. It builds on three earlier studies from this laboratory, on foldability of lattice proteins, on the relationship between folding and function, and on elastic network models (GNM and ANM) of real proteins, which established hierarchical modularity but left its physical nature and modelling consequences open. Using statistical mechanics (exactly solvable hierarchical foldon models, lattice simulations, residue-level folding ensembles and harmonic elastic networks), an information-theoretic approach (tree decomposition of total correlation, mutual information and exact free-energy cycles) and unsupervised learning (variational autoencoders), and calibrating every index against compact-polymer nulls, matched alternative trees, optimized flat partitions and a frozen protocol on reserved protein families, we find that the hierarchy of proteins is mechanical and organizational rather than thermodynamic. Native contact networks have reproducible, sequence-local nested modules, but nesting itself is shared with compact polymers. What is specific to proteins is how units are grouped and where softness is placed: the contact-derived grouping captures binding-induced motion better than groupings matched in size and compactness (also in 18 reserved families), relative motions of nested modules form a stiffness ladder, mirrored by an information ladder in the exact tree decomposition of total correlation, whose top-level joint is softer and less informative and whose lower pairs are more cohesive than in alternative trees, the softness of that joint is carried by the

Authors

Institutions

Publication Details

Journal
ChemRxiv
Published
2026-11-17
DOI
https://doi.org/10.26434/chemrxiv.15012337/v1
Primary Topic
Protein Structure and Dynamics
Type
preprint
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

The Nature of Protein Hierarchies and Implications for Hierarchical Modelling: An AI-Assisted Computational Study with Statistical Mechanics, Information Theory and Unsupervised Learning

Shubin Liu
1 links
ChemRxiv
Protein Structure and Dynamics
preprint

The Nature of Protein Hierarchies and Implications for Hierarchical Modelling: An AI-Assisted Computational Study with Statistical Mechanics, Information Theory and Unsupervised Learning

Shubin Liu
preprint en

Abstract

Proteins are widely described as hierarchies of nested structural units, and hierarchical models are widely used to simulate them, yet what kind of hierarchy proteins actually possess, and what that hierarchy implies for how they should be modelled, has not been established. This work addresses these two questions directly. It builds on three earlier studies from this laboratory, on foldability of lattice proteins, on the relationship between folding and function, and on elastic network models (GNM and ANM) of real proteins, which established hierarchical modularity but left its physical nature and modelling consequences open. Using statistical mechanics (exactly solvable hierarchical foldon models, lattice simulations, residue-level folding ensembles and harmonic elastic networks), an information-theoretic approach (tree decomposition of total correlation, mutual information and exact free-energy cycles) and unsupervised learning (variational autoencoders), and calibrating every index against compact-polymer nulls, matched alternative trees, optimized flat partitions and a frozen protocol on reserved protein families, we find that the hierarchy of proteins is mechanical and organizational rather than thermodynamic. Native contact networks have reproducible, sequence-local nested modules, but nesting itself is shared with compact polymers. What is specific to proteins is how units are grouped and where softness is placed: the contact-derived grouping captures binding-induced motion better than groupings matched in size and compactness (also in 18 reserved families), relative motions of nested modules form a stiffness ladder, mirrored by an information ladder in the exact tree decomposition of total correlation, whose top-level joint is softer and less informative and whose lower pairs are more cohesive than in alternative trees, the softness of that joint is carried by the

ChemRxiv
University of North Carolina at Chapel Hill (US)
Protein Structure and Dynamics
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon