Recognizing Local and Global Structural Motifs at the Atomic Scale

Piero Gasparotto, Robert H. Meißner, Michele CeriottiJournal of Chemical Theory and Computation 2018 · DOI · arXiv

Abstract

Most of the current understanding of structure-property relations at the molecular and the supramolecular scales can be formulated in terms of the stability of and the interactions between a limited number of recurring structural motifs (e.g., H-bonds, coordination polyhedra, and protein secondary structure). Here we demonstrate an algorithm to automatically recognize such patterns, based on the identification of local maxima in the probability distributions observed in atomistic computer simulations, which is robust to the dimensionality and the sparsity of the reference atomistic data. We first discuss its main features, demonstrating some on artificial data sets, and then show how it can be applied to identify coordination environments in Lennard-Jones clusters and to recognize secondary-structure patterns in the simulation of an oligopeptide. To assess the applicability of this algorithm for motifs that involve several interdependent degrees of freedom, we also employ it to identify groups of conformers of the cluster and the polypeptide, considered in their entirety. The motifs identified by analyzing atomistic simulations can be used to interpret and rationalize the stability and behavior of the system at hand, and also as a tool to accelerate sampling, in association with biased molecular dynamics schemes.

Introduction

In the field of molecular dynamics (MD), researchers are currently facing a data deluge. Modern supercomputing capabilities allow for the simulation of systems containing millions of atoms over microsecond timescales. However, the interpretation of these massive trajectories remains a significant bottleneck. Traditionally, scientists have relied on "order parameters"—manually designed geometric rules—to identify structural motifs like hydrogen bonds or protein folds.

This trial-and-error approach is becoming increasingly impractical as systems grow in complexity. Furthermore, existing automated clustering methods often struggle with high-dimensional data, sparse sampling, or non-standard geometries (such as periodic dihedral angles). This paper introduces a refined Probabilistic Analysis of Molecular Motifs (PAMM) framework designed to be robust, unbiased, and capable of recognizing both local environments and global configurations without requiring human-defined rules for the motifs.

The PAMM Framework

The proposed workflow improves upon the original PAMM scheme through several key innovations. Initially, atoms are described using "fingerprints" that respect physical symmetries like rotations and translations. To manage computational costs, the algorithm employs a greedy farthest-point sampling to create a sparse grid over the data.

A core innovation is the use of Adaptive Kernel Density Estimation (KDE). The algorithm estimates probability density using a Gaussian kernel with a local bandwidth that adapts to the data's density and dimensionality. This prevents the over-smoothing common in simpler methods. Following density estimation, Quick-Shift clustering identifies "modes" (local maxima) in the probability distribution, where each maximum represents a stable structural motif.

Figure 4: Model fitting using Gaussian/Von Mises Mixture Models to identify distinct probability modes (motifs α, β, and γ).

Once modes are identified, they are fitted to a Gaussian Mixture Model (GMM) to create Probabilistic Motif Identifiers (PMIs). These act as fuzzy indicators of the probability that a specific configuration belongs to a given motif. To handle noise and non-Gaussian shapes, the authors utilize bootstrapping to test cluster stability. Unstable or overlapping clusters are merged hierarchically, creating a dendrogram that reflects the underlying free-energy landscape.

Figure 1: The overlap matrix and resulting dendrogram illustrate the hierarchical merging of motifs based on statistical bootstrapping.

Applications and Key Results

The authors validated the PAMM approach across three distinct datasets:

  • Artificial 2D Datasets: Used to demonstrate how meta-clustering successfully groups non-Gaussian shapes, such as concentric rings, which typically confuse standard algorithms.
  • Lennard-Jones Clusters ($LJ_{38}$): In this classic chemical physics benchmark, the algorithm identified 8 distinct local motifs (e.g., core vs. surface atoms) and successfully separated global structural states like truncated octahedrons from liquid-like configurations.
  • GB1 Oligopeptide: For this 16-residue peptide, PAMM partitioned the Ramachandran plot into regions corresponding to known secondary structures (alpha-helices, beta-sheets) while also identifying non-traditional motifs in unstructured regions.

Discussion and Future Directions

A significant insight from this work is that the hierarchical merging of clusters qualitatively resembles a "disconnectivity graph." The distance between clusters in the stability matrix correlates with the free-energy barriers between those states. By performing clustering in high-dimensional space, the authors can identify distinct states that might appear merged in 2D projections.

While the algorithm is powerful, the number of clusters can still be sensitive to initial localization parameters. However, the hierarchical merging step significantly mitigates this issue. Looking forward, these PMIs could serve as "collective variables" for biased molecular dynamics. By automatically identifying stable states, researchers can use the algorithm to accelerate the sampling of rare events in complex biological and chemical systems.

Summary updated

Abstract

Most of the current understanding of structure-property relations at the molecular and the supramolecular scales can be formulated in terms of the stability of and the interactions between a limited number of recurring structural motifs (e.g., H-bonds, coordination polyhedra, and protein secondary structure). Here we demonstrate an algorithm to automatically recognize such patterns, based on the identification of local maxima in the probability distributions observed in atomistic computer simulations, which is robust to the dimensionality and the sparsity of the reference atomistic data. We first discuss its main features, demonstrating some on artificial data sets, and then show how it can be applied to identify coordination environments in Lennard-Jones clusters and to recognize secondary-structure patterns in the simulation of an oligopeptide. To assess the applicability of this algorithm for motifs that involve several interdependent degrees of freedom, we also employ it to identify groups of conformers of the cluster and the polypeptide, considered in their entirety. The motifs identified by analyzing atomistic simulations can be used to interpret and rationalize the stability and behavior of the system at hand, and also as a tool to accelerate sampling, in association with biased molecular dynamics schemes.
BibTeX
@article{gasparotto2018recognizing,
  author = {Piero Gasparotto and Robert H. Meißner and Michele Ceriotti},
  title = {{Recognizing Local and Global Structural Motifs at the Atomic Scale}},
  journal = {Journal of Chemical Theory and Computation},
  year = {2018},
  doi = {10.1021/acs.jctc.7b00993},
}
Figure