NEWS.md

In MosaiClusteR: An Umbrella Framework for Multi-Source and Multi-Omics Clustering

MosaiClusteR 0.1.4

• CharacteristicFeatures() bug fix. When called with binData (the common case, as in the documented example) an inner loop index overwrote the cluster index, so only the first cluster's characteristic features were returned, and a datanames mis-index left the per-source result list unnamed. Both are fixed: features are now computed and named for every cluster.
• plot_pathways(style = "circular"). New radial (sunburst-style) bar plot that groups the enriched terms by cluster around a central hub. Bar length encodes \eqn{-\log_{10}} p and colour encodes the cluster; each term is a single-line radial label that stays legible as the number of terms grows. Complements the existing "dotplot" and "lollipop" styles.
• enriched_features() - new helper that turns [CharacteristicFeatures()] output into a per-cluster list of compact term/pvalue/Count tables (keeping the significant, enriched features and tidying their names), ready to hand straight to plot_pathways().
• plot_pathways(style = "circular") gained cluster_cols to supply an explicit per-cluster palette (e.g. to match the colours used by Cyclogram()), and wrap to control radial label wrapping.
• FeaturesOfCluster() and BinFeaturesPlot_SingleData() gained cexB, cexR and cexN arguments to control the object, feature and name label sizes; Cyclogram() now defaults to a larger cex.labels.

MosaiClusteR 0.1.2

• Distance(distmeasure = "euclidean") now computes a true Euclidean distance (via stats::dist). Previously it routed to FD::gowdis (Gower), which returned NaN when a feature was constant across the selected objects, producing NA distance matrices. Gower is still available via distmeasure = "gower".
• ComparePlot() (and Cyclogram()) now build a default colour palette when cols is not supplied. Previously the default cols = NULL caused an error ("number of items to replace is not a multiple of replacement length") from ColorsNames(), so a palette had to be passed explicitly.
• ComparePlot() now checks up front that all results describe the same objects and raises a clear message otherwise, instead of a cryptic "missing value where TRUE/FALSE needed" error. (Results from different data sets / sample sizes cannot be compared.)

MosaiClusteR 0.1.1

• M_ABCpp() renamed to M_ABC(). The multi-source Aggregating Bundles of Clusters function is now M_ABC(); its arguments, behaviour and return value are unchanged. Update any calls from M_ABCpp(...) to M_ABC(...).
• Metadata and documentation: the "MoSaIC" name now expands to Multi-Omics Similarity Aggregation and Integrative Clustering in R, the package licence is CC BY-SA 4.0, and the four-layer framework overview in the README is now a scalable SVG figure (man/figures/framework.svg) that reads on both light and dark backgrounds.

MosaiClusteR 0.1.0

First public release. MosaiClusteR is an umbrella framework for multi-source and multi-omics clustering: it gathers a broad catalogue of integration methods behind one consistent interface, together with the tooling needed to preprocess, select, evaluate, compare, characterise and biologically interpret the results.

Framework

• Unified interface. Every method takes the same input — a List of numeric matrices over the same objects (objects in rows) — and returns the same {DistM, Clust} contract, so methods are interchangeable and directly comparable. Precomputed dissimilarities are accepted via type = "dist".
• Four-layer workflow. Preprocess → baseline → integrate → evaluate.

Integration methods (five paradigms)

• Direct: ADC(), ADEC().
• Similarity-based: WeightedClust(), WonM(), SNF(), NEMO().
• Graph-based: EnsembleClustering() (CSPA/HGPA/MCLA), HBGF(), ClusteringAggregation().
• Voting / consensus: CEC(), CVAA(), ConsensusClustering(), EvidenceAccumulation(), LinkBasedClustering() (CTS/SRS/ASRS), and the Aggregating Bundles of Clusters family M_ABC() / M_ABCdist() / M_ABCdist.WC() / M_ABCdeep() (a deep-learning, autoencoder-based variant).
• Hierarchy-based: HierarchicalEnsembleClustering(), EHC().
• Factor / low-rank / spectral: intNMF(), spectral_clustering(), LUCID() (supervised, outcome-guided).
• Cluster() provides the single-source baseline primitive.

Preprocessing, selection and evaluation

• Preprocessing: Normalization() and Distance() (Euclidean, Manhattan, Tanimoto/Jaccard for binary data, correlation-based, and more).
• Cluster-number selection: SelectnrClusters(), ChooseCluster().
• Evaluation: mosaic_labels(), cluster_agreement() (ARI / NMI / Jaccard / purity), compare_clusterings(), CompareSilCluster(), CompareSvsM(), and data-driven source weighting (DetermineWeight_SilClust(), DetermineWeight_SimClust()).

Feature weighting and data nuggets

• A data-nugget compression provides a robust, big-data-friendly feature weighting as an alternative to variance: create_data_nuggets(), nugget_feature_weights(), nugget_cluster(), Wkmeans(), Whclust().

Characterisation, visualisation and interpretation

• Characterisation / navigation: CharacteristicFeatures(), FeatSelection(), FeaturesOfCluster(), FindCluster(), FindElement(), TrackCluster(), ReorderToReference(), SharedComps(), SimilarityMeasure().
• Visualisation: ComparePlot(), ClusterPlot(), Cyclogram(), SimilarityHeatmap(), HeatmapPlot(), ProfilePlot(), ContFeaturesPlot(), BinFeaturesPlot_SingleData()/_MultipleData(), BoxPlotDistance(), and colour helpers (ColorPalette(), ColorsNames(), ClusterCols(), LabelCols()).
• Biological interpretation (omics): DiffGenes(), PathwayAnalysis(), Pathways(), Geneset.intersect(), SharedGenesPathsFeat() and related functions connect clusters to genes and pathways.

Data and tests

• Bundled mosaic_toy dataset and the mosaic_sim() structured-data generator.
• A testthat suite covering distances, weighting schemes, the integration methods, evaluation metrics and data nuggets.

Notes

• A compiled kernel (src/) accelerates the consensus step; building from source needs a C/C++ toolchain (Rtools on Windows). Heavier and specialised dependencies — SNFtool, circlize, plotrix, ggplot2, igraph, limma, MLP, biomaRt, org.Hs.eg.db and others — are declared under Suggests and loaded only when the corresponding method is called.
• A few methods rely on an external graph-partitioning executable (e.g. METIS for EnsembleClustering(), the METIS path of EHC(), and LinkBasedClustering(linkBasedMethod = "asrs")) and raise an informative error when it is absent; pure-R alternatives (HBGF(), EvidenceAccumulation(), CTS/SRS) cover the same need.

MosaiClusteR documentation built on Oct. 25, 2026, 1:08 a.m.