Multivariate analysis rework, dendrogram/sample-correlation QC tools, and cleanup#8
Merged
Conversation
- goto_search now tells the user to run an analysis first instead of silently doing nothing when the search tab is opened before self.analysisrun is set, closing the "add runcheck before searching" TODO. - Removed main.py's unused PyQt5/stdlib/groupsets imports (platform, GroupSet, several never-referenced Qt classes), verified via pyflakes + grep cross-check; no behavior change, 130 existing tests still pass. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Lays the Qt-free foundation for the "groups can be reordered" TODO: GroupSetModel.move(from_index, to_index) reorders groupsets and keeps the selection on the moved/shifted item by identity (not GroupSet's value-based __eq__, since two freshly-added default groupsets compare equal). 8 new tests in test_groupsets.py. UI wiring (drag-and-drop on listWidget_pltgrps) intentionally left for later -- it would need to be verified against a live GUI session to confirm it interacts correctly with updatesets()'s existing blockSignals dance, which isn't something to ship unverified. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
PCA itself already exists (plot_PCA/goto_pca/checkbox field) -- the remaining gap is specifically loadings/biplot visualization of which features drive each component, which plot_PCA doesn't do yet. Reworded so the TODO doesn't read as if PCA support is still missing. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
New code/ordination.py: PCA, NMDS (the metric-MDS-warm-started non-metric MDS already used by the soon-to-be-renamed plot_PCA, kept verbatim), and PLS-DA, plus a Qt-free port of the data-loading/ technical-replicate-collapsing logic the plot currently has hardcoded off (plotting.py: `parent.collapsereps = False#...`). The collapse-replicate logic is a near-verbatim port of the original, not a rewrite -- its header-juggling (round-tripping through a CSV to relabel an unstack() result) is easy to get subtly wrong by inspection alone, so it's preserved as-is and verified empirically instead: test_ordination.py constructs a synthetic peak table with 3 samples across 2 biological groups, 3 technical-replicate injections each, and asserts collapsing lands on exactly 3 rows (one per Sample) -- not 9 (uncollapsed) and not 2 (would mean biological replicates got merged too). Cross-checked against real example data with a scratch script (27 injections / 9 samples / 3 groups -> collapses to 9, not 3). Also caught and fixed a real bug while validating against real data: PLSRegression's default scale=True standardizes X internally, so the original explained-variance-ratio calc (component score variance / unscaled total variance) silently produced ratios around 1e-6 instead of the ~0.7 a well-separated dataset should show. Fixed with scale=False, matching PCA's plain-centered treatment. OPLS-DA is intentionally not implemented (see ordination.py's module docstring) -- no scikit-learn support, and the alternatives (an unmaintained third-party package, or a from-scratch implementation with no reference dataset to validate against) are both riskier than shipping PCA/NMDS/PLS-DA now and revisiting OPLS-DA later. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
plot_PCA only ever ran NMDS (with a PCA rotation applied afterward purely to orient axes) -- renamed to plot_ordination and reworked to genuinely support PCA, NMDS, and PLS-DA, selectable via a combo-box switcher bar inserted above the plot canvas (same runtime widget-substitution pattern as searchtree.py's filter bar), plus a Scores/Loadings view toggle. The math moved to the new Qt-free ordination.py (previous commit); this is the Qt plumbing on top of it. - Axis labels now show percent-variance-explained where meaningful (PCA/PLS-DA: real feature-space variance; NMDS: labeled distinctly as embedding variance, since it isn't the same quantity). - Loadings view shows the top-25 features by vector magnitude as origin-anchored arrows (thousands of features can't all be drawn legibly) -- but whichever feature is currently highlighted elsewhere in the app is always included regardless of magnitude, via a second pre-created highlight artist (plot_ordination.highlight_loading(), called from MainWindow._refresh_highlight()) following the same convention every other plot's highlight marker already uses. - Restored "Collapse Technical Replicates": plotting.py previously had this hardcoded off (`parent.collapsereps = False#...isChecked()`); now reads the real checkbox via ordination.load_ordination_matrix(). - checkBox_pca's visible text/btn_pca's tooltip changed from "PCA" to "Multivariate" -- the underlying objectName/analysis_params.PCA attribute are unchanged for .mpct save-file compatibility. - Verified the view/method-switching lifecycle (Scores<->Loadings, method changes, the highlight-on-demand path for a feature outside the default top-25) against real example data with an offscreen Qt harness before considering this done -- in particular confirmed ui_plot.reset()'s mpl_disconnect(self.event) doesn't error when switching away from Scores view (where the pick-event connection lives) and back. devnotes.md documents all of the above plus the OPLS-DA deferral (unmaintained pyopls package, or a from-scratch implementation with no reference dataset to validate against -- both riskier than shipping PCA/NMDS/PLS-DA now). Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
…tyling All four issues caught only by checking against real data / the live GUI, not by inspection: - PCA/PLS-DA now autoscale features (mean-center + unit-variance) before fitting. Raw mass-spec intensities span a huge range across features (confirmed: feature std devs from ~1.8 to ~10,000 on real example data) -- without scaling, a handful of high-abundance features dominated both explained-variance and loadings, which is why loadings were showing up "in the thousands" while most were tiny, and why %explained looked unusually high. NMDS is deliberately left unscaled (Bray-Curtis dissimilarity is conventionally computed on raw/relative abundances). - NMDS axis labels no longer show percent-explained at all -- it's a rank-based embedding, not a linear decomposition, so it doesn't canonically have that quantity the way PCA/PLS-DA do. Shows stress (the conventional NMDS fit-quality metric) as the plot title instead. - Loadings-view axis limits are now set explicitly from the actually- plotted data: ax.annotate()'s arrows don't reliably drive matplotlib's autoscale the way ax.scatter()/ax.plot() do (confirmed empirically -- plotted points could fall outside the auto-picked view), which is what required manually rescaling each axis before. Also fixed top_loadings() being called against the full (up to 10-component) loadings instead of just the 2 displayed ones, which could let an irrelevant-to-this-view feature crowd out a genuinely prominent one. - Switcher bar: capped to a fixed max height so it doesn't eat canvas space, and restyled for page_pca's light background (white combo boxes, dark text) instead of searchtree.py's dark-theme styling, which was the wrong context here. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
New clusterpurity.py colors dendrogram branches green wherever a whole group's leaves merge together before meeting any other group, plus a Technical/Biological Replicates switcher on the dendrogram tab (mirrors plot_ordination's method/view bar) and a plot-title purity summary (n_pure/n_total). Applies to both the regular and bootstrap (PvClust) dendrogram paths. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Purity coloring now uses red (not black) for branches that mix more than one group, and a new "Color: None" mode reproduces the tab's pre-purity-coloring appearance exactly (plain black, no title) -- fixes a regression where dropping color_threshold=0 made "None" fall back to scipy's default multi-color palette instead of plain black. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
gen_treemap/gen_upsetplt used to savefig() a PNG to the repo root and load it into a QLabel via QPixmap -- no zoom/pan/save toolbar, and a flat raster rewritten on every run. plot_treemap/plot_upset now draw directly onto a persistent FigureCanvas, wired into _generate_plots() via the same _create_or_reset pattern every other plot uses, so they regenerate on both a fresh analysis and the Apply button (previously only on a fresh analysis). Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
…xes to per-plot bars - clusterpurity.purity_link_color_func now distinguishes pure (green), bridge (red -- the specific merge where a new group first meets an existing one), and neutral (black -- combining two already-impure clades, no new information). Previously every ancestor of a single mixing event also rendered red, painting most of the tree's upper structure red regardless of how localized the actual mixing was. - "Bootstrap Analysis" and "Collapse Technical Replicates" moved off the global plot-config dialog (where each only ever affected one plot) onto that plot's own switcher bar: plot_dendrogram gets a "Bootstrap" checkbox, plot_ordination gets a "Collapse Replicates" checkbox. The now-orphaned dialog widgets are hidden at runtime (not edited out of the generated ui_plotparam.py); 'bootstrap' is dropped from paramfields.CHECKBOX_FIELDS since it's no longer pickled, consistent with the dendrogram/ordination tabs' other per-session-only view state. - Delete code/treemap.png and code/test_upsetplt.png: dead tracked files left over from before the canvas-based rendering change -- nothing reads or writes them anymore. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
The previous "bridge vs neutral" heuristic still mis-colored real data: it could mark a high-level merge red just because one side was a single freshly-introduced pure clade, and could miss genuine tangles where two already-impure children share a label without either side being trivially pure. purity_link_color_func now classifies each merge by comparing the two children's label sets directly: disjoint sets (no label in common) -> neutral/black, a clean join even if one side is impure from an unrelated tangle further down; overlapping sets -> red, definitive proof some label's leaves are split across this exact merge. Verified against the real dataset's bootstrap dendrogram: only the actual scattered-replicate merges render red, and every higher-level merge joining that region with cleanly-resolved samples stays black. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- ordination.replicate_label_components() numbers each Injection's biological and technical replicate rank (BioRep#/TechRep#) within its Biolgroup/Sample, unconditionally (works fine when either count is 1). A new "Use Sample/Group Names" checkbox in the dendrogram's switcher bar swaps the raw file/injection names for <Biolgroup>_b<BioRep#>_s<TechRep#> (or _b<BioRep#> alone in the Biological Replicates view) -- useful when the real file names are long or uninformative. - pvclust.plot_dendrogram's AU/BP annotations used a fixed icoord-unit x-shift that shrank to an ever-smaller pixel gap as leaf count grew (icoord-to-pixel ratio shrinks with more leaves in the same plot width), eventually merging "AU"/"BP" into illegible overlapping text. Fixed with ax.annotate(..., textcoords='offset points', ha='right'/'left'), which keeps a constant pixel gap regardless of leaf count, plus leaf-count-scaled fontsize. Also removed a plt.figure()/plt.tight_layout() pair that created and abandoned an unused Figure on every redraw. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
…ements - multivariate.md: rewrite from NMDS-only to cover the full PCA/NMDS/PLS-DA ordination tab — method switcher, scores/loadings view, collapse-replicates checkbox, stress metric for NMDS, %explained for PCA/PLS-DA. - group-analysis.md: document all four dendrogram switcher-bar controls (View, Color, Bootstrap, Use Sample/Group Names). - changelog.md: add 2026 entries for ordination rework, dendrogram purity coloring/switchers/label options, AU/BP annotation fix, and canvas-plot UpSet/treemap. - development.md: add ordination.py, clusterpurity.py, csvcache.py to the hand-written-code list; add ordination to the test-coverage list. - index.md: expand the "mid-2026 updates" note to mention ordination and dendrogram reworks. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
sklearn not added to tests, caused ci build test failure
robertsamples
added a commit
that referenced
this pull request
Jun 30, 2026
… and cleanup (#8) * Reword and reprioritize main.py TODO list * Add search-tab run-check, clean up main.py's dead imports - goto_search now tells the user to run an analysis first instead of silently doing nothing when the search tab is opened before self.analysisrun is set, closing the "add runcheck before searching" TODO. - Removed main.py's unused PyQt5/stdlib/groupsets imports (platform, GroupSet, several never-referenced Qt classes), verified via pyflakes + grep cross-check; no behavior change, 130 existing tests still pass. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com> * Add GroupSetModel.move() for groupset reordering (model layer only) Lays the Qt-free foundation for the "groups can be reordered" TODO: GroupSetModel.move(from_index, to_index) reorders groupsets and keeps the selection on the moved/shifted item by identity (not GroupSet's value-based __eq__, since two freshly-added default groupsets compare equal). 8 new tests in test_groupsets.py. UI wiring (drag-and-drop on listWidget_pltgrps) intentionally left for later -- it would need to be verified against a live GUI session to confirm it interacts correctly with updatesets()'s existing blockSignals dance, which isn't something to ship unverified. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com> * Clarify stale PCA TODO note PCA itself already exists (plot_PCA/goto_pca/checkbox field) -- the remaining gap is specifically loadings/biplot visualization of which features drive each component, which plot_PCA doesn't do yet. Reworded so the TODO doesn't read as if PCA support is still missing. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com> * Add Qt-free multivariate ordination backend (PCA/NMDS/PLS-DA) New code/ordination.py: PCA, NMDS (the metric-MDS-warm-started non-metric MDS already used by the soon-to-be-renamed plot_PCA, kept verbatim), and PLS-DA, plus a Qt-free port of the data-loading/ technical-replicate-collapsing logic the plot currently has hardcoded off (plotting.py: `parent.collapsereps = False#...`). The collapse-replicate logic is a near-verbatim port of the original, not a rewrite -- its header-juggling (round-tripping through a CSV to relabel an unstack() result) is easy to get subtly wrong by inspection alone, so it's preserved as-is and verified empirically instead: test_ordination.py constructs a synthetic peak table with 3 samples across 2 biological groups, 3 technical-replicate injections each, and asserts collapsing lands on exactly 3 rows (one per Sample) -- not 9 (uncollapsed) and not 2 (would mean biological replicates got merged too). Cross-checked against real example data with a scratch script (27 injections / 9 samples / 3 groups -> collapses to 9, not 3). Also caught and fixed a real bug while validating against real data: PLSRegression's default scale=True standardizes X internally, so the original explained-variance-ratio calc (component score variance / unscaled total variance) silently produced ratios around 1e-6 instead of the ~0.7 a well-separated dataset should show. Fixed with scale=False, matching PCA's plain-centered treatment. OPLS-DA is intentionally not implemented (see ordination.py's module docstring) -- no scikit-learn support, and the alternatives (an unmaintained third-party package, or a from-scratch implementation with no reference dataset to validate against) are both riskier than shipping PCA/NMDS/PLS-DA now and revisiting OPLS-DA later. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com> * Rework the mislabeled "PCA" plot into a multivariate ordination tab plot_PCA only ever ran NMDS (with a PCA rotation applied afterward purely to orient axes) -- renamed to plot_ordination and reworked to genuinely support PCA, NMDS, and PLS-DA, selectable via a combo-box switcher bar inserted above the plot canvas (same runtime widget-substitution pattern as searchtree.py's filter bar), plus a Scores/Loadings view toggle. The math moved to the new Qt-free ordination.py (previous commit); this is the Qt plumbing on top of it. - Axis labels now show percent-variance-explained where meaningful (PCA/PLS-DA: real feature-space variance; NMDS: labeled distinctly as embedding variance, since it isn't the same quantity). - Loadings view shows the top-25 features by vector magnitude as origin-anchored arrows (thousands of features can't all be drawn legibly) -- but whichever feature is currently highlighted elsewhere in the app is always included regardless of magnitude, via a second pre-created highlight artist (plot_ordination.highlight_loading(), called from MainWindow._refresh_highlight()) following the same convention every other plot's highlight marker already uses. - Restored "Collapse Technical Replicates": plotting.py previously had this hardcoded off (`parent.collapsereps = False#...isChecked()`); now reads the real checkbox via ordination.load_ordination_matrix(). - checkBox_pca's visible text/btn_pca's tooltip changed from "PCA" to "Multivariate" -- the underlying objectName/analysis_params.PCA attribute are unchanged for .mpct save-file compatibility. - Verified the view/method-switching lifecycle (Scores<->Loadings, method changes, the highlight-on-demand path for a feature outside the default top-25) against real example data with an offscreen Qt harness before considering this done -- in particular confirmed ui_plot.reset()'s mpl_disconnect(self.event) doesn't error when switching away from Scores view (where the pick-event connection lives) and back. devnotes.md documents all of the above plus the OPLS-DA deferral (unmaintained pyopls package, or a from-scratch implementation with no reference dataset to validate against -- both riskier than shipping PCA/NMDS/PLS-DA now). Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com> * Fix ordination feedback: scaling, axis limits, NMDS %explained, bar styling All four issues caught only by checking against real data / the live GUI, not by inspection: - PCA/PLS-DA now autoscale features (mean-center + unit-variance) before fitting. Raw mass-spec intensities span a huge range across features (confirmed: feature std devs from ~1.8 to ~10,000 on real example data) -- without scaling, a handful of high-abundance features dominated both explained-variance and loadings, which is why loadings were showing up "in the thousands" while most were tiny, and why %explained looked unusually high. NMDS is deliberately left unscaled (Bray-Curtis dissimilarity is conventionally computed on raw/relative abundances). - NMDS axis labels no longer show percent-explained at all -- it's a rank-based embedding, not a linear decomposition, so it doesn't canonically have that quantity the way PCA/PLS-DA do. Shows stress (the conventional NMDS fit-quality metric) as the plot title instead. - Loadings-view axis limits are now set explicitly from the actually- plotted data: ax.annotate()'s arrows don't reliably drive matplotlib's autoscale the way ax.scatter()/ax.plot() do (confirmed empirically -- plotted points could fall outside the auto-picked view), which is what required manually rescaling each axis before. Also fixed top_loadings() being called against the full (up to 10-component) loadings instead of just the 2 displayed ones, which could let an irrelevant-to-this-view feature crowd out a genuinely prominent one. - Switcher bar: capped to a fixed max height so it doesn't eat canvas space, and restyled for page_pca's light background (white combo boxes, dark text) instead of searchtree.py's dark-theme styling, which was the wrong context here. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com> * Add dendrogram purity coloring: technical/biological replicate QC view New clusterpurity.py colors dendrogram branches green wherever a whole group's leaves merge together before meeting any other group, plus a Technical/Biological Replicates switcher on the dendrogram tab (mirrors plot_ordination's method/view bar) and a plot-title purity summary (n_pure/n_total). Applies to both the regular and bootstrap (PvClust) dendrogram paths. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com> * Dendrogram: polyphyletic branches in red, add a no-coloring option Purity coloring now uses red (not black) for branches that mix more than one group, and a new "Color: None" mode reproduces the tab's pre-purity-coloring appearance exactly (plain black, no title) -- fixes a regression where dropping color_threshold=0 made "None" fall back to scipy's default multi-color palette instead of plain black. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com> * Replace treemap/upset PNG round-trip with real canvas plots gen_treemap/gen_upsetplt used to savefig() a PNG to the repo root and load it into a QLabel via QPixmap -- no zoom/pan/save toolbar, and a flat raster rewritten on every run. plot_treemap/plot_upset now draw directly onto a persistent FigureCanvas, wired into _generate_plots() via the same _create_or_reset pattern every other plot uses, so they regenerate on both a fresh analysis and the Apply button (previously only on a fresh analysis). Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com> * Dendrogram: bridge-only red coloring; move bootstrap/collapse checkboxes to per-plot bars - clusterpurity.purity_link_color_func now distinguishes pure (green), bridge (red -- the specific merge where a new group first meets an existing one), and neutral (black -- combining two already-impure clades, no new information). Previously every ancestor of a single mixing event also rendered red, painting most of the tree's upper structure red regardless of how localized the actual mixing was. - "Bootstrap Analysis" and "Collapse Technical Replicates" moved off the global plot-config dialog (where each only ever affected one plot) onto that plot's own switcher bar: plot_dendrogram gets a "Bootstrap" checkbox, plot_ordination gets a "Collapse Replicates" checkbox. The now-orphaned dialog widgets are hidden at runtime (not edited out of the generated ui_plotparam.py); 'bootstrap' is dropped from paramfields.CHECKBOX_FIELDS since it's no longer pickled, consistent with the dendrogram/ordination tabs' other per-session-only view state. - Delete code/treemap.png and code/test_upsetplt.png: dead tracked files left over from before the canvas-based rendering change -- nothing reads or writes them anymore. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com> * Fix dendrogram coloring: red = proven non-monophyly (label-set overlap) The previous "bridge vs neutral" heuristic still mis-colored real data: it could mark a high-level merge red just because one side was a single freshly-introduced pure clade, and could miss genuine tangles where two already-impure children share a label without either side being trivially pure. purity_link_color_func now classifies each merge by comparing the two children's label sets directly: disjoint sets (no label in common) -> neutral/black, a clean join even if one side is impure from an unrelated tangle further down; overlapping sets -> red, definitive proof some label's leaves are split across this exact merge. Verified against the real dataset's bootstrap dendrogram: only the actual scattered-replicate merges render red, and every higher-level merge joining that region with cleanly-resolved samples stays black. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com> * Dendrogram: add Use Sample/Group Names labels; fix AU/BP label scaling - ordination.replicate_label_components() numbers each Injection's biological and technical replicate rank (BioRep#/TechRep#) within its Biolgroup/Sample, unconditionally (works fine when either count is 1). A new "Use Sample/Group Names" checkbox in the dendrogram's switcher bar swaps the raw file/injection names for <Biolgroup>_b<BioRep#>_s<TechRep#> (or _b<BioRep#> alone in the Biological Replicates view) -- useful when the real file names are long or uninformative. - pvclust.plot_dendrogram's AU/BP annotations used a fixed icoord-unit x-shift that shrank to an ever-smaller pixel gap as leaf count grew (icoord-to-pixel ratio shrinks with more leaves in the same plot width), eventually merging "AU"/"BP" into illegible overlapping text. Fixed with ax.annotate(..., textcoords='offset points', ha='right'/'left'), which keeps a constant pixel gap regardless of leaf count, plus leaf-count-scaled fontsize. Also removed a plt.figure()/plt.tight_layout() pair that created and abandoned an unused Figure on every redraw. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com> * Docs: update mkdocs guide for ordination rework and dendrogram improvements - multivariate.md: rewrite from NMDS-only to cover the full PCA/NMDS/PLS-DA ordination tab — method switcher, scores/loadings view, collapse-replicates checkbox, stress metric for NMDS, %explained for PCA/PLS-DA. - group-analysis.md: document all four dendrogram switcher-bar controls (View, Color, Bootstrap, Use Sample/Group Names). - changelog.md: add 2026 entries for ordination rework, dendrogram purity coloring/switchers/label options, AU/BP annotation fix, and canvas-plot UpSet/treemap. - development.md: add ordination.py, clusterpurity.py, csvcache.py to the hand-written-code list; add ordination to the test-coverage list. - index.md: expand the "mid-2026 updates" note to mention ordination and dendrogram reworks. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com> * correlation matrix control improvements * Update tests.yml sklearn not added to tests, caused ci build test failure --------- Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
This branch bundles several related improvements that grew out of working through the main.py TODO list.
##Multivariate Analysis tab rework
##Dendrogram quality-control coloring
##Sample Correlation Matrix
##UpSet plot and treemap
##Other cleanup
Compatibility: No changes to the saved .mpct file format; old save files still load correctly.
Testing: All 159 headless unit tests pass; GUI changes verified manually against example data.