Modern biology increasingly relies on large-scale screening to generate high dimensional datasets with potential to accelerate discovery. However, analysing these complex datasets remains challenging, due to hierarchical biological structure, uncertainty in the true number of groups, and high dimensional noise that confounds genuine biological signal. Here we present an unsupervised consensus clustering tool, Untangled, that aggregates clustering solutions across granularities to construct a stability-based representation, followed by cluster number optimization and systematic evaluation of cluster robustness. Through extensive benchmarking on simulated datasets, ground truth datasets and diverse high-dimensional screening applications, we demonstrate that Untangled reliably recovers underlying relationships, resolves stable meaningful substructure, and more effectively prioritizes robust clusters with shared biological mechanisms and conserved phenotypic responses than alternative clustering approaches. These results establish Untangled as a scalable framework for cluster discovery to guide efficient follow-up investigation from high-dimensional biological datasets.

MOTIVATION: The advent of long-read DNA sequencing is allowing complete assembly of highly repetitive genomic regions for the first time, including the megabase-scale satellite repeat arrays found in many eukaryotic centromeres. The assembly of such repetitive regions creates a need for their de novo annotation, including patterns of higher order repetition. To annotate tandem repeats, methods are required that can be widely applied to diverse genome sequences, without prior knowledge of monomer sequences. RESULTS: Tandem Repeat Annotation and Structural Hierarchy (TRASH) is a tool that identifies and maps tandem repeats in nucleotide sequence, without prior knowledge of repeat composition. TRASH analyses a fasta assembly file, identifies regions occupied by repeats and then precisely maps them and their higher order structures. To demonstrate the applicability and scalability of TRASH for centromere research, we apply our method to the recently published Col-CEN genome of Arabidopsis thaliana and the complete human CHM13 genome. AVAILABILITY AND IMPLEMENTATION: TRASH is freely available at:https://github.com/vlothec/TRASH and supported on Linux.