Changes in version 0.99.2 Bug fixes - A network with Ensembl transcript IDs is always annotated from Ensembl 107, or from an EnsDb passed as transcript_biotypes; a transcript_biotype vertex attribute or a transcript_biotypes table overrides it for the transcripts it lists. Before, a table or attribute covering every vertex, or include_nmd = TRUE with include_nsd = TRUE, skipped the annotation: decay transcripts outside the network were not blocked as seeds and could be remapped to a coding representative, and the unfiltered bundled network had no biotypes or gene symbols. If the annotation cannot be loaded, the session is still created when both flags are TRUE (without biotypes and gene symbols) or when every vertex has a biotype (with a warning that decay transcripts outside the network cannot be blocked). - A network without Ensembl transcript IDs does not load the Ensembl 107 annotation when the session is created; an unannotated one gives the "unknown transcript biotypes" error directly. The annotation field of the eligibility audit names the sources that were used. - IsoNetAnalyzeR() merges repeated edges of a supplied network using the pi of the predicted copies only, as for the bundled network. - get_metadata(obj, "analysis_biotype_exclusions")[[label]] is a zero-row table when no seed was excluded, instead of missing. - Rerunning an analysis label removes the seed-support and duplicate-resolution audits of the earlier run. - n_genes_dropped_opposite_collision in the coverage report counts only genes dropped for protein-identical seeds in opposite directions, not genes removed by seed_padj_col or without mapped seeds. - pct_up_found and pct_down_found divide by the resolved representatives, so protein-identical seeds count once. - Versioned Ensembl gene IDs (ENSG00000070371.16) are accepted in the gene_id of run_differential_pulldown(), and by aggregate_duplicate_counts(), which keeps them in its output, and resolve_duplicate_isoforms(). The help no longer says that gene symbols work in gene_id. - distance_from_seed is the distance to the nearest seed. It was the mean over seeds, which was Inf for every node when one seed could not reach another. - summarize_subgraphs() returns one row per graph for unnamed, partly named and duplicate-named lists; an unnamed list gave no rows. - aggregate_duplicate_counts() uses feature_data for a SummarizedExperiment. Without it, an ID column missing from rowData is taken from the row names (genes of mapped transcripts from the map), the rowData ID and other columns are kept, and a list-valued gene_id, as in tximeta objects, gives each row its first gene. It returns the map provenance, which was NULL. - run_component_ora() stores the p-values and adjusted p-values of fgsea::fora() unrounded; p-values below 5e-7 were stored as 0. - run_fgsea() uses fgseaMultilevel unless nperm is given, so p-values are no longer limited to 1/1001 and fgsea no longer warns that fgseaSimple is used. The method is recorded in the result's params. - Gene sets (named lists or GeneSetCollections) are matched to the ranked genes in run_fgsea() and to the network genes in run_component_ora(): IDs that are such genes are kept, and other Ensembl gene IDs and gene symbols match every such gene with or without a version suffix. Collections without an identifier type, as read by GSEABase::getGmt(), are accepted: their IDs are taken as Ensembl gene IDs when all look like Ensembl gene IDs, and as gene symbols otherwise. An empty collection, or gene sets with no matching gene, give a clear error. - The MSigDB GO gene sets are loaded and converted once per R session instead of on every run_fgsea() or run_component_ora() call. - Versioned identifiers of an EnsDb passed as transcript_biotypes match the network's transcripts and genes. - When the gene-symbol annotation cannot be loaded, plots and printed summaries no longer retry loading it every time. - Up is green and down orange in every plot and key; component and cross-gene plots used orange for up and sky blue for down. Box plots draw genes without an up/down difference in grey. - The "x", "y", "up" and "down" label strategies of the cross-gene plots label the genes with the highest values of their own statistic; "up" and "down" take only positive and negative differences, and several strategies take turns in the order given, so each is represented. - get_isoform_stats() flags the seeds of results saved without an is_seed column from the gene's seed isoforms; the earlier fallback, distance_from_seed == 0, missed seeds of multi-seed genes. - plot_component() lists "none" when an analysis has no component analyses. A missing (NA) analysis or component label, and group_by_gene = NA, give the intended argument errors, and the warning for an analysis with an empty cross-gene summary names the analysis. Other changes - The all_edges argument of plot_network() is deprecated and has no effect. It changed the drawn network only through an undocumented pi_score edge attribute, which the bundled network does not have. - future and future.apply are suggested instead of imported; they are needed only for parallel = TRUE. - Gene-level component analyses and plots build the merged graph of 1,000 genes in about a second instead of several minutes, and run_differential_pulldown() is about eight times faster on the bundled network (1,000 genes in about half a minute instead of over four minutes). Changes in version 0.99.1 Bundled network - The source table lists six transcript pairs twice and five transcripts paired with themselves. These entries counted twice in the vertex degrees behind the neighbourhood cutoff and changed the pulldowns that reached them; in the GSE141432 example, the up-labelled APAF1 neighbourhood included a DIABLO transcript only because of them. Each repeated pair is now one edge with the mean pi of its copies, and the self-pairs are removed (158,447 edges). Supplied networks are simplified the same way when a session is created. Bioconductor classes and resources - Result tables are S4Vectors DataFrames, both in the object slots and as returned by the accessors. Table inputs can be a data.frame or a DataFrame. - Transcript biotypes and gene symbols come from Ensembl release 107, the GENCODE v41 gene build of the bundled network, through AnnotationHub record AH104864, instead of tables shipped with the package. Decay exclusion is unchanged. isonet_transcript_biotypes() and isonet_gene_names() accept an EnsDb and return DataFrames, and the transcript_biotypes argument of IsoNetAnalyzeR() accepts an EnsDb. - Gene sets can be a GSEABase GeneSetCollection; gene symbols are mapped to Ensembl gene IDs. Without pathways, run_fgsea() and run_component_ora() fetch the MSigDB 2023.1 GO gene sets with the msigdb package instead of msigdbr. The examples and the vignette take the MSigDB Hallmark collection from msigdb instead of a file shipped with the package. - isonet_duplicate_map() returns a DataFrame, with the build provenance in metadata(). Changes in version 0.99.0 Version prepared for submission to Bioconductor. Bundled network - The bundled network now contains the 158,458 predicted interaction edges (interact == 1) and all 58,164 transcripts, instead of all 2,853,266 scored pairs. Every analysis already restricted the network to these edges, so pulldowns, lookups, components, enrichment and diagnostics are unchanged; this was verified against 0.6.5. Pulldowns are about ten times faster and the source package is under Bioconductor's 5 MB limit. - Counts of all scored edges change accordingly: the session edge count, the eligibility audit fields edges_before/edges_after, and the diagnostics field full_edges now count predicted interaction edges. plot_network(obj, all_edges = TRUE) on a whole session no longer draws non-interacting pairs. - inst/scripts/reduce_network.R derives the new network from the 0.6.5 file; inst/PROVENANCE.md documents both checksums. New features - get_metadata() returns session metadata such as coverage, seed-support and eligibility audits, replacing direct slot access. - get_pulldown_graph() returns a gene's up- or down-seeded pulldown graph. - get_network() returns the session network. - Network plots label nodes with gene symbols. plot_network() and plot_component() gain node_labels = c("symbol", "id", "none") to switch to transcript or gene identifiers or to hide labels, and gene plot titles show the symbol next to the gene ID. Gene labels in the cross-gene scatter and boxplots also use symbols. - isonet_gene_names() returns the GENCODE v41 gene symbols of the bundled network (inst/extdata/gene_names_gencode_v41.csv.gz, produced by inst/scripts/make_gene_names.R), for annotating result tables. Printed output - Printed summaries are aligned in columns. - Session creation reports counts with thousands separators and states that excluded NMD and NSD transcripts take their edges with them, which explains the two edge counts. - "Access with" hints use the name of the variable passed as obj, and the enrichment summaries show "user-supplied" instead of NA for gene sets passed through pathways. Plot fixes - The direction = "overlay" plot of plot_network() draws down-only edges dashed and shared edges thicker, has a short centred title with the gene symbol, and puts the legends below the plot. - Plot titles, axis labels and legends use only characters that R's standard PDF device can draw, so plots saved as PDF keep all their text. The cross-gene boxplots label up-minus-down values "difference (nodes)" and similar instead of using the Greek capital delta. - Plots of a neighbourhood with only two nodes no longer warn "data length [3] is not a sub-multiple or multiple of the number of rows [2]". - plot_component(colour_by = "direction") draws one legend for colour and one for seed status instead of two legends both titled "Direction"; equidistant neighbours are labelled "Equidistant". - The plot_cross_gene_overview() subtitle no longer describes the up/down comparison as a change between conditions. - The gene-connectivity panel of plot_network_diagnostics() leaves room for the count above its tallest bar. Documentation and data - New vignette, "Introduction to IsoNetAnalyzeR", analysing isoform switches between obese type 2 diabetic and lean adipose tissue (GEO GSE141432). - Example data: gse141432_t2d_vs_lean_seeds.csv (137 seeds of 65 genes) and the MSigDB Hallmark gene sets (msigdb_hallmark_ensembl.rds), with generating scripts in inst/scripts/. - inst/PROVENANCE.md describes how the network was derived from STRING physical interactions and attributes STRING's CC BY 4.0 licence. - Every exported function and class has runnable examples. - The help page for run_network_diagnostics() no longer truncates the description of the default sticky-node rule. Package infrastructure - biocViews, URL and BugReports added; ORCID iDs, copyright holders and the thesis supervisors listed in Authors@R; licence declared as MIT + file LICENSE with the full text in LICENSE.md. - Requires R 4.6.0 or later. - msigdbr moved to Suggests; withr added to Imports to draw the diagnostics plot subsample with a local seed. Changes in version 0.6.5 - Add a concise README with direct user-guide links, a self-contained installed instructions file, and a separate network-eligibility explanation with references. - Keep the complete MIT licence in a single LICENSE file; update R licence metadata and the README link while preserving the copyright notice and terms. - Consolidate the default NMD/NSD exclusions from 0.5.5 and the optional retained-significant-seed validation from 0.6.0 into the 0.6.5 release. These scientific rules and the pinned reference resources are unchanged from the checked 0.6.0 analysis archive. - Separate focal-seed role from neighbourhood membership in differential overlay plots (ISO-PLOT-001). Focal seeds now use a reddish-purple fill (#CC79A7) with directional seed shapes; non-seed nodes and edges use green/up, orange/down and blue/both-direction colours. This corrects the plotting behaviour that was not included in the earlier 0.6.0 archive. - Add an executable retained-support example to the vignette and full workflow, clarify nonsignificant complementary seeds and the exploratory three-column interface, and update installation and validation instructions for 0.6.5. - Preserve the raw embedded network, protein-identical transcript map and GENCODE v41 biotype table. Existing checksum-pinned study archives and RNA results remain separate from this release. Changes in version 0.6.0 - Add a retained-seed support gate to run_differential_pulldown() via seed_padj_col and seed_fdr. After decay filtering, ID resolution and mapping, every retained gene must contain at least one significant retained seed and at least one up and one down seed. Non-significant complementary seeds remain allowed; there is no seed maximum. - Store gene/row support audits in metadata$analysis_seed_support. An excluded, unmapped or collapsed-away transcript cannot supply significance for the final seed set. - RNA testing and effect-size selection remain upstream. The GSE141432 v0.6 analysis enables the gate explicitly; three-column exploratory inputs remain compatible when seed_padj_col = NULL. Changes in version 0.5.5 - New sessions exclude annotated NMD and NSD transcripts from protein networks by default. Independent include_nmd/include_nsd flags restore either class explicitly. - Pinned transcript-biotype annotations and exclusion audits distinguish biological eligibility from statistical seed selection and network coverage. - Custom networks require known transcript biotypes when exclusions are enabled. Existing saved sessions are unchanged; rebuild them to apply the new policy. - No maximum seed count is imposed by the package; upstream studies should declare effect/significance rules and whether both directions are required. Changes in version 0.5.0 Protein-identical isoform support - Ship a provenance-checked, transcriptome-wide map from 16,630 redundant transcript identifiers to 9,760 protein representatives. The complete map is available through isonet_duplicate_map(); 10,592 mappings are applicable to the embedded network and 6,038 additional mappings remain available for upstream count aggregation. - Add resolve_duplicate_isoforms() for version normalization and graph-compatible duplicate resolution. Differential pulldown, direct pulldown, and isoform lookup resolve identifiers by default and retain row-level audits. Set resolve_duplicates = FALSE for the exact identifier behaviour used by 0.4.0. - Detect pseudo-switches created when oppositely directed submitted isoforms resolve to the same protein representative. Differential analysis drops the affected gene before graph expansion; same-direction collisions collapse to one representative. - Distinguish cleaned submitted counts from resolved-representative counts. Dominance scores now explicitly use the resolved analysis table. Count aggregation and validation - Add aggregate_duplicate_counts() for matrices, data frames, and optional single-assay SummarizedExperiment inputs. It preserves nonmapped features, feature order, gene and library totals, and emits a complete arithmetic and mapping audit. - Expand coverage reports to show both resolution paths and network outcomes, with submitted and unique-representative totals reported separately. - Add installed-resource, custom-graph, collision, audit-persistence, count arithmetic, overflow, and optional-container regression tests. The embedded R/sysdata.rda network is unchanged. Changes in version 0.4.0 Usability and validation - Add a persistent three-state network coverage summary: unmapped, mapped without a predicted edge, and mapped with a predicted edge, overall and per supplied direction. Mapping below 10% now emits diagnostic warnings. - Validate non-finite and fractional integer arguments, stored cutoffs, input duplicates and conflicts, transcript-to-gene mappings, and the complete custom-network contract. - Support networks with no predicted edges and report complete per-gene processing failure clearly while preserving valid unmapped, isolated, and one-direction results. - Correct partially missing multi-seed lookups. - Return plot objects visibly without explicit printing, enabling assignment and headless ggsave() without creating Rplots.pdf. Documentation - Define the package as analysis-agnostic and keep seed derivation and publication-specific analyses outside the package. - Replace proxy-direction examples with a caller-prepared synthetic workflow, qualify compositional dIF guidance, and describe results as static predicted seed neighbourhoods. - Document reproducible enrichment using seeded runs and frozen pathway lists. Changes in version 0.3.2 Correctness fixes - Preserve mapped isolated seed isoforms and genes through cross-gene graph construction, component analysis, component plotting, single-direction plotting, and differential overlay plotting. A mapped seed with no retained interactions is now represented as a valid one-node/zero-edge result. - Correct gene-level component bookkeeping so ENSG vertices are compared with gene identifiers rather than ENST seed identifiers when counting direction contributions. - Plot a stored component using its authoritative node-table membership, so component IDs remain correct when component filters were used. - Store explicit per-node seed flags in virtual-pulldown statistics, retaining a legacy fallback for serialized objects created before version 0.3.2. - Give singleton summaries finite structural conventions (density, average degree, and clustering equal to zero; centralization remains undefined). Documentation and maintenance - Correct package author/maintainer metadata, Rd markup, and tidy-evaluation global-variable declarations. - Transfer release maintenance to Dimitrios S. Kanakoglou while retaining Nikolas Mumm as the original package author. - Correct the GitHub installation URL, vignette name, and cross-gene summary column count in the README and vignette. - Add regression and end-to-end tests covering gene/isoform component counts, isolated seeds, explicit filters, plotting, serialization, invalid public inputs, fully unmapped analyses, typed empty outputs, and single-direction sparse analyses. - Run plotting regressions on an explicit temporary graphics device so package tests cannot leave Rplots.pdf in the source tree. - Replace soft-deprecated igraph plotting aliases with their current API names; singleton and filtered plotting now complete without deprecation warnings and retain identical graph semantics.