HiMaLAYAS: enrichment-based annotation and visualization of hierarchically clustered matrices
Abstract
Summary: Hierarchical clustering reveals structure in high-dimensional biological matrices, but the resulting dendrogram-defined clusters are commonly visualized rather than tested for annotation enrichment. Existing workflows often require separate tools to cluster data, test enrichment, and display heatmaps. Here, we introduce Hierarchical Matrix Layout and Annotation Software (HiMaLAYAS), a Python framework for post hoc enrichment-based annotation and visualization of hierarchically clustered matrices. HiMaLAYAS treats dendrogram-defined clusters as statistical units, tests categorical annotations for enrichment, controls multiple testing, and renders significant annotations alongside clusters in a single workflow. Applied to a Saccharomyces cerevisiae genetic interaction profile similarity matrix, HiMaLAYAS annotated biological-process enrichment at parent-cluster and zoomed-cluster levels. Parent-level annotations persisted across clustering-parameter sweeps, exceeded null expectations from randomized cluster membership and annotation labels, and remained stable under matrix perturbation. HiMaLAYAS also annotated clusters in a non-biological country-sector input-output matrix, extending its application beyond biology. Availability and Implementation: HiMaLAYAS is a Python package available via pip, distributed under the BSD 3-Clause License at https://github.com/himalayas-base/himalayas, and archived on Zenodo at https://doi.org/10.5281/zenodo.18610373.
02 /Specification & Classification
Click any tag to browse matching papers- Citations
- 0
- Organism
- Research Focus
- Approach
In-depth summary
Claude Opus 5.5 · xhigh effort · 8 Nov 2026 · from the full text of version 3 (25 Sept 2026) · CC BY 4.0
HiMaLAYAS, a Python package, cuts dendrograms, tests clusters for categorical annotation enrichment (hypergeometric, Benjamini-Hochberg FDR), and renders significant annotations beside clustered heatmaps in one workflow.
Dendrogram-defined clusters in heatmap analyses are usually inspected by eye instead of being statistically tested for annotation enrichment, and clustering, enrichment and plotting tend to live in separate packages. HiMaLAYAS joins these steps and checks whether its annotations hold up under clustering-parameter changes, randomized null models and matrix noise.
Key results
ShownOn a yeast genetic interaction similarity matrix, parent-level clusters were labeled with distinct biological processes. Re-running the analysis on one zoomed parent cluster split it into finer functional subclusters while keeping the same enrichment background.
Costanzo et al. (2016) matrix restricted to 1,053 high-profile-variance genes; GO BP enrichment at FDR ≤ 0.05. Parent-level terms included vesicle-mediated transport, mRNA splicing and DNA replication. Within the zoomed mRNA-splicing cluster, subclusters were labeled mRNA transport, RNA splicing and chromatin organization. · 3.1 Annotating the Saccharomyces cerevisiae…; Figure legends
ShownThe observed enrichment signal was far stronger than two randomization null models: one shuffled cluster membership and the other shuffled annotation labels.
The parent-level analysis gave 331 significant cluster-term pairs (q ≤ 0.05). The largest null value was 5, so the observed count was more than 60-fold higher. Empirical P = 0.001 for both 1,000-replicate nulls (same-size random clusters with annotations fixed; label permutations with clusters fixed). · 2.3 Robustness and null-analysis design; 3.2 Testing annotation robustness; Figure legends
ShownHeadline GO BP annotations were mostly preserved after the yeast matrix was perturbed with symmetric zero-mean Gaussian noise and reclustered.
Noise was added at 0.05, 0.10, 0.20, 0.30 and 0.50 of the off-diagonal standard deviation, with five replicates per fraction. Across the perturbed runs, a headline term was the top-ranked annotation in 138 of 175 and still passed the significance cutoff in 174 of 175. · 2.3 Robustness and null-analysis design; 3.2 Testing annotation robustness
SuggestedThe seven headline parent-level terms were mostly recovered across sweeps of dendrogram distance threshold and minimum cluster size, suggesting the annotations do not depend on a single cut or size setting.
Reference settings were distance threshold 16 and minimum cluster size 30. Recovery was scored in the cluster with the largest gene overlap with each reference cluster. The main text describes recovery only qualitatively and gives no counts; the curves are in Fig. 2A,B. · 2.3 Robustness and null-analysis design; 3.2 Testing annotation robustness; Figure legends
SuggestedOutside biology, HiMaLAYAS clustered origin country-sectors of the 2014 World Input-Output Database by their intermediate-flow profiles and found clusters enriched for shared origin countries.
Shown only in Supplementary Fig. S4. The main text gives no enrichment counts, q-values or cluster numbers for this matrix. · 3.3 A single workflow for annotating and…
Methods, data and reproducibility
Users provide a real-valued matrix and categorical term-to-observation annotations. HiMaLAYAS clusters the matrix hierarchically (linkage, metric and distance threshold are configurable), merges or excludes undersized clusters, runs one-sided hypergeometric tests per cluster-term pair with Benjamini-Hochberg FDR, and renders the top-ranked terms in matrix or condensed hierarchy views.
- Costanzo et al. (2016) yeast genetic interaction profile similarity matrix (Pearson correlation of profiles), subset to high-profile-variance genes · Saccharomyces cerevisiae · n = 1,053 genes
- Gene Ontology Biological Process term-to-gene memberships used as the categorical annotations for enrichment testing · Saccharomyces cerevisiae
- Zoomed view of the parent cluster enriched for mRNA splicing, reanalyzed against the full-matrix enrichment background · Saccharomyces cerevisiae
- 2014 World Input-Output Database country-sector matrix of intermediate flows, annotated by origin country
- Methods & tools
- HiMaLAYAS (Python package), One-sided hypergeometric test, Benjamini-Hochberg false discovery rate procedure, Gene Ontology Biological Process (GO BP), simplifyEnrichment (suggested for grouping redundant GO terms), WGCNA (compared), clusterProfiler (compared), ComplexHeatmap (compared), VisHiC (compared), funcExplorer (compared), Clustergrammer (compared), Jupyter notebook tutorials
- Code
- https://github.com/himalayas-base/himalayashttps://github.com/himalayas-base/himalayas-docshttps://doi.org/10.5281/zenodo.18610373
- Data & accessions
- https://github.com/himalayas-base/himalayas-publication
- Software & environment
- Python package installed via pip and distributed under the BSD 3-Clause License, with a Zenodo archive. No package or dependency versions are stated.
- Reproducibility gaps
- No hardware, runtime or memory is reported. The main text gives no source accessions for the Costanzo et al. (2016) matrix or the World Input-Output Database matrix; it says only that datasets and figure workflows are in the publication repository.
Limitations (4)
- HiMaLAYAS does not filter out semantically redundant GO terms before visualization; users who want such terms grouped can turn to separate tools such as simplifyEnrichment. · stated by the authors
- Annotations depend on how clusters are defined: the dendrogram cut, the minimum cluster size and matrix noise can all change cluster composition and therefore which terms are enriched. · stated by the authors
- The robustness tests (parameter sweeps, null models, noise perturbation) cover only the parent-level analysis of one yeast dataset; the zoomed-cluster and World Input-Output analyses were not tested for robustness. · noted in this summary
- The comparison with WGCNA, clusterProfiler, ComplexHeatmap and other tools is a qualitative feature summary. There is no quantitative benchmark of enrichment results, runtime or scalability against these tools. · noted in this summary
Who should read:Read this if you annotate clustered heatmaps (for example, genetic interaction or expression similarity matrices) and want dendrogram clusters tested for term enrichment and plotted with FDR-controlled labels in Python.
Reproducibility ledger
Nobody has reported trying to reproduce this paper yet. Did you replicate an experiment or run the code? Your report saves the next lab a week.
Sign in to file a reproducibility report.
Related papers
- Interactive Visual Exploration And Refinement Of Cluster Assignments
Michael Kern, Alexander Lex, … Chris R. Johnson·BMC BioinformaticsPeer-reviewed·Sep 12, 2017
- simplifyEnrichment: an R/Bioconductor package for Clustering and Visualizing Functional Enrichment Results
Zuguang Gu, Daniel Hübschmann·Genomics Proteomics & BioinformaticsPeer-reviewed·Jun 6, 2022
- clusterProfiler: An universal enrichment tool for functional and comparative study
Tianzhi Wu, Erqiang Hu, … Guangchuang Yu·The InnovationPeer-reviewed·Jul 1, 2021
- SummArIzeR: Simplifying cross-database enrichment result clustering and annotation via large language models
Marie Brinkmann, Michael Markus Bonelli, Anela Tosevska·BioinformaticsPeer-reviewed·Feb 27, 2026
- Association Plots: Visualizing associations in high-dimensional correspondence analysis biplots
Elzbieta Gralinska, Martin Vingron·bioRxiv·Oct 23, 2020
- HiPiler: Visual Exploration Of Large Genome Interaction Matrices With Interactive Small Multiples
Fritz Lekschas, Benjamin Bach, … Hanspeter Pfister·IEEE Transactions on Visualization and Computer GraphicsPeer-reviewed·Aug 28, 2017