How to Use Kluster

Kluster is a structural clustering workflow for organizing many protein models by pairwise shape similarity rather than by sequence identity. It combines structure alignment, dimensionality reduction, and density-based clustering so researchers can discover dominant conformational states, structural outliers, or recurring fold families inside a large ensemble.

On Neurosnap, users upload multiple Input Structures and choose the Alignment Tool, Features, Method, and Dimensions that define how similarity is measured and projected. The workflow is well suited to mutant panels, AlphaFold or molecular dynamics ensembles, and collections of homologous proteins where a simple sorted TM-score list is too limited.

The results are intentionally visual. A cluster projection reveals broad state organization, while a similarity-ordered animation helps users see structural transitions or recurring motifs across the dataset.

Kluster first computes pairwise structural relationships with either TM-align or US-align, using TM-score, RMSD, or both as the feature space for comparison. Those pairwise metrics are then projected into two or three dimensions with UMAP, t-SNE, or PCA, and DBSCAN is used to identify clusters without forcing the user to predefine the number of states.

On Neurosnap, Features changes whether clustering leans more on global fold similarity, absolute coordinate deviation, or a combination of both. Dimensions affects whether the main projection is easier to read in 2D or preserves more geometric nuance in 3D, while Animation Projection Algorithm controls how the ensemble is ordered when generating a one-dimensional structural walk-through.

Researchers typically use Kluster to identify cluster representatives, structural outliers, and possible transition series before moving into more detailed structural alignment, annotation, or experimental interpretation.

• Many-structure workflow: Kluster is designed for ensembles and panels, not only one-against-one structural comparison.
• Flexible similarity definition: TM-score, RMSD, or both can be used depending on whether the question is fold similarity, coordinate deviation, or a mix of the two.
• Projection control: Alignment, reduction, dimensionality, and animation settings let researchers tailor the visual summary to the dataset.
• State and outlier discovery: The cluster view is useful for finding conformational families, rare structures, and representative models quickly.