CLoNe: automated clustering based on local density neighborhoods for application to biomolecular structural ensembles

Abstract

Motivation Proteins are intrinsically dynamic entities. Flexibility sampling methods, such as molecular dynamics or those arising from integrative modeling strategies are now commonplace and enable the study of molecular conformational landscapes in many contexts. Resulting structural ensembles increase in size as technological and algorithmic advancements take place, making their analysis increasingly demanding. In this regard, cluster analysis remains a go-to approach for their classification. However, many state-of-the-art algorithms are restricted to specific cluster properties. Combined with tedious parameter fine-tuning, cluster analysis of protein structural ensembles suffers from the lack of a generally applicable and easy to use clustering scheme. Results We present CLoNe, an original Python-based clustering scheme that builds on the Density Peaks algorithm of Rodriguez and Laio. CLoNe relies on a probabilistic analysis of local density distributions derived from nearest neighbors to find relevant clusters regardless of cluster shape, size, distribution and amount. We show its capabilities on many toy datasets with properties otherwise dividing state-of-the-art approaches and improves on the original algorithm in key aspects. Applied to structural ensembles, CLoNe was able to extract meaningful conformations from membrane binding events and ligand-binding pocket opening as well as identify dominant dimerization motifs or inter-domain organization. CLoNe additionally saves clusters as individual trajectories for further analysis and provides scripts for automated use with molecular visualization software. Availability www.epfl.ch/labs/lbm/resources, github.com/LBM-EPFL/CLoNe Supplementary information Supplementary data are available at Bioinformatics online.

CATBOSS: Cluster Analysis of Trajectories Based on Segment Splitting

“…While the algorithm is fairly robust to cutoff choice, a list-position based cutoff may present issues with clusters of varying densities. 35 In order to include information from all data points, while minimizing user input, for all segment-based clustering trials, the kernel density estimator cutoff was set to the average distance to the ln( N )-th nearest neighbor, where N is the number of trajectory segments considered. This choice of cutoff was motivated by the idea that the number of nearest neighbors k ( N ) must adapt to the underlying data distribution as the number of samples N → ∞.…”

“…In the original implementation, this value is set to a fixed (second) percentile of the sorted list of pairwise distances. While the algorithm is fairly robust to cutoff choice, a list-position based cutoff may present issues with clusters of varying densities . In order to include information from all data points, while minimizing user input, for all segment-based clustering trials, the kernel density estimator cutoff was set to the average distance to the ln­( N )-th nearest neighbor, where N is the number of trajectory segments considered.…”

“… 28 − 34 Limitations of this method previously identified by the scientific community include the need for the user to specify the cutoff distance for the kernel density estimator and the need for the user to visually inspect the generated decision graph and manually assign cluster centroids as well as quadratic memory complexity. 35 , 36 The last issue in particular can make memory requirements for a typical MD data set balloon to hundreds of gigabytes, necessitating the use of expensive high-end hardware. This problem can be mitigated by recomputing the pairwise distances as needed, rather than storing them (which trades memory for computational complexity) or using local approximations for density estimation.…”

“… 36 Later implementations of the method may be run on large data sets on regular desktop machines. 37 Several other groups have also proposed extensions of the method that address the aforementioned shortcomings; 35 , 36 however, none have, to our knowledge, entirely eliminated user input or reduced memory complexity without computational trade-off or the use of approximations.…”

“…This method relies on the observation that cluster centroids exhibit a relatively high local density compared to their neighbors and a large distance from any points of higher density. This method has proven competent at handling clusters of varying shapes, sizes, and densities and has already been applied effectively to MD data sets. − Limitations of this method previously identified by the scientific community include the need for the user to specify the cutoff distance for the kernel density estimator and the need for the user to visually inspect the generated decision graph and manually assign cluster centroids as well as quadratic memory complexity. , The last issue in particular can make memory requirements for a typical MD data set balloon to hundreds of gigabytes, necessitating the use of expensive high-end hardware. This problem can be mitigated by recomputing the pairwise distances as needed, rather than storing them (which trades memory for computational complexity) or using local approximations for density estimation .…”