Multiscale Persistent Functions for Biomolecular Structure Characterization

Kelin Xia, Zhiming Li, Lin Mu

TL;DR

This paper introduces multiscale persistent functions, a novel framework that combines topological data analysis (persistent homology) with multiscale rigidity functions to characterize biomolecular structures. By transforming discrete structural data into a continuous density map and analyzing its topological "lifespan," the authors provide a robust way to calculate conformational entropy and introduce a new metric, the Protein Structure Index (PSI), to quantify structural regularity.

What Problem Does This Paper Address?

Understanding how proteins fold and interact requires a precise way to measure their conformational entropy—essentially, how "disordered" or "flexible" a molecule is. Traditionally, researchers calculate this by plotting backbone angles (like the Ramachandran plot) and dividing the graph into a grid.

However, this "grid-based" approach has a major flaw: the resulting entropy value is extremely sensitive to the size of the grid. If the grid is too fine, every data point looks unique; if it’s too coarse, distinct structural features are lumped together. There is no "natural" scale. This paper seeks to move beyond arbitrary grids by using topology to let the data define its own natural clusters and scales.

Methods

The authors propose a multi-step pipeline to characterize structure:

  1. Coarse-Grained Representation: Instead of looking at every atom, they use $C_\alpha$ atoms and calculate two specific angles: the pseudo-bond angle ($\theta$) and the virtual dihedral angle ($\tau$).
  2. Multiscale Rigidity Functions: They transform these discrete $(\theta, \tau)$ points into a continuous "density map" using kernel functions (like Gaussian or Lorentz kernels). A resolution parameter ($\eta$) acts as a "lens," allowing them to view the data at different levels of detail.
  3. Persistent Homology (PH): They perform a "density filtration" on this map. Imagine a rising water level: as the level changes, "islands" (clusters of data) appear and eventually merge. PH tracks the "birth" and "death" of these features, recording them in a barcode.
  4. Persistent Entropy: Using the lengths of these barcodes, they calculate a multiscale persistent entropy. Unlike traditional methods, this approach is data-driven and captures the intrinsic topological features of the protein's configuration space.
  5. Protein Structure Index (PSI): They define PSI based on the $\beta_0$ (connected components) persistent entropy to specifically measure the "regularity" of a protein.

Datasets

The study utilizes two primary datasets:

  • 110 High-Resolution Proteins: A curated set of proteins with resolutions $\leq 1.5$ Å, used to test the PSI and compare the new method against traditional grid-based entropy.
  • SCOPe Database Subset: Approximately 900 proteins categorized into all-alpha, all-beta, and mixed alpha-beta classes to validate the method's classification power.

Key Insights and Ideas

  • Topology as a Natural Classifier: Instead of forcing data into a grid, persistent homology treats clusters of bond angles as topological features. This allows the method to identify "regular" structures (like $\alpha$-helices) because their angles are tightly clustered, creating long-lasting (persistent) features in the barcode.
  • The "Mediate Range" Advantage: The authors found that while their method matches traditional results at extreme scales (very fine or very coarse), it performs significantly better in the "middle" resolutions. This is where traditional methods lose structural nuance, but persistent functions preserve the underlying "shape" of the data.
  • Quantifying "Regularity": The introduction of the Protein Structure Index (PSI) provides a formal way to say a protein is "orderly." A low PSI indicates a highly regular structure (like a perfect helix), while a high PSI indicates a messy, "irregular" structure dominated by loops and disordered regions.

Key Results

  • Protein Classification: The method achieved a clear separation between all-alpha and all-beta proteins using only dihedral and pseudo-bond angle information. Mixed alpha-beta proteins consistently fell into the middle region, as expected.
  • Robustness: The persistent entropy showed a high correlation with traditional methods at high resolutions (PCC > 0.98) but remained more stable and informative at lower resolutions where traditional methods fail.
  • PSI Validation: In the test of 110 proteins, the most "regular" structure (lowest PSI) was a regular $\beta$-sheet (PDB ID: 1GK7), while the most "irregular" (highest PSI) was a protein with large portions of loops and disordered regions (PDB ID: 3ZFP).

Limitations and Future Work

  • Higher-Dimensional Topology: This paper primarily focuses on $\beta_0$ (connected components/clusters). The authors acknowledge that more complex information could be extracted from $\beta_1$ (loops) and $\beta_2$ (voids) topological entropies, which they plan to explore in future work.
  • Computational Complexity: While more robust than a grid, persistent homology is computationally more intensive. Scaling this to massive molecular dynamics trajectories might require further algorithmic optimization.
  • Application to Other Systems: While tested on proteins, the authors suggest PSI is a general metric that could describe "regularity" in any complex system, though these non-biological applications were not explored here.

Summary updated

Abstract

In this paper, we introduce multiscale persistent functions for biomolecular structure characterization. The essential idea is to combine our multiscale rigidity functions with persistent homology analysis, so as to construct a series of multiscale persistent functions, particularly multiscale persistent entropies, for structure characterization. To clarify the fundamental idea of our method, the multiscale persistent entropy model is discussed in great detail. Mathematically, unlike the previous persistent entropy or topological entropy, a special resolution parameter is incorporated into our model. Various scales can be achieved by tuning its value. Physically, our multiscale persistent entropy can be used in conformation entropy evaluation. More specifically, it is found that our method incorporates in it a natural classification scheme. This is achieved through a density filtration of a multiscale rigidity function built from bond and/or dihedral angle distributions. To further validate our model, a systematical comparison with the traditional entropy evaluation model is done. It is found that our model is able to preserve the intrinsic topological features of biomolecular data much better than traditional approaches, particularly for resolutions in the mediate range. Moreover, our method can be successfully used in protein classification. For a test database with around nine hundred proteins, a clear separation between all-alpha and all-beta proteins can be achieved, using only the dihedral and pseudo-bond angle information. Finally, a special protein structure index (PSI) is proposed, for the first time, to describe the "regularity" of protein structures. Essentially, PSI can be used to describe the "regularity" information in any systems.
BibTeX
@article{xia2017multiscale,
  author = {Kelin Xia and Zhiming Li and Lin Mu},
  title = {{Multiscale Persistent Functions for Biomolecular Structure Characterization}},
  journal = {Bulletin},
  year = {2017},
  doi = {10.1007/s11538-017-0362-6},
}

Loading related content...

Figure