NEWS

---

KODAMA 0.99.6

• Added RunFastPCA() as the primary native float32 PCA interface for numeric matrices, SingleCellExperiment, SpatialExperiment, Seurat, and Giotto. Container methods select and transpose expression assays internally and store PCA scores using each framework's dimensional-reduction interface. The lower-level kodama_pca() and KODAMA.pca() names remain available for compatibility.
• Added a session-wide options(n.cores = ...) setting, with N_CORES as an environment fallback, for PCA, graph construction, KODAMA optimization, and visualization. CPU-only spatial feature selection follows the same core-count setting. Explicit n.cores arguments retain priority.
• Preserved sparse assays during spatial feature selection and bounded dense C++ work buffers by feature batches.
• Spatial feature-selection container methods now return the updated object, store statistics and ranks in feature metadata, and allow RunFastPCA(nfeatures = ...) to consume the ranking directly.

KODAMA 0.99.2

• Standardized session-wide backend selection across KODAMA, fastPLS, fastEmbedR, and faissR through options(backend = ...) and BACKEND. Legacy KODAMA-specific selectors remain compatibility fallbacks.

KODAMA 0.99.1

• Established tkcaccia/KODAMA as the canonical repository for the new C++-backed R package and preserved the classic implementation in tkcaccia/KODAMAlegacy, aligning the repository and package names required for Bioconductor submission.
• Replaced the legacy R implementation with bindings to the standalone float32 kodama-cpp library.
• Added reusable KODAMA.graph(), KNN and PLS-LDA optimization, CPU, CUDA, and Metal execution, and graph-input workflows.
• Added adapters for SingleCellExperiment, SpatialExperiment, Seurat, and Giotto objects.
• Retained the historical MetRef, USA, and lymphoma datasets.
• Renamed the former implementation repository to KODAMAlegacy; this package now owns the canonical KODAMA name.