# 12 Exports and Graphs ## Quick Answer EVd3x exports both data tables and graph artifacts. Use the full analysis ZIP for statistics and reproducibility, then use per-graph PNG/SVG exports for figures. Interactive views may be capped for speed, but full exports preserve comprehensive datasets where available. ## What this does Defines how each graph is computed, what it encodes, and how to export both the plotted view and underlying tabular data. ## Inputs - API payloads from node, pathway, communication, disease, LR, and specificity modules - Active filters (source, p-value cutoff, direction, detail cap, selected cells) - Graph registry keys in frontend (`graphRegistry`) ## Outputs - Full analysis CSV ZIP (`Export Full Analysis Bundle`) - Graph image ZIP (`Export Graphs Bundle`) - Per-graph popup viewer with fit/reset, drag/pan, PNG/SVG download - Right-panel overlays (collapsible legend and PPI tree) for graph-first interpretation ## How calculated Graphs render from analysis payloads and stateful filters. SVG charts are serialized from DOM (`svg`) and Cytoscape graphs are exported via Cytoscape image methods. For dense views, UI may apply caps; manifest metadata records truncation and limits. ## What to download Minimum manuscript package: - `evd3x_export__.zip` - `evd3x_graphs__.zip` - `99_meta/export_manifest.csv` - `99_meta/export_warnings.csv` when present ## Known limits - Very dense networks can reduce readability in-browser; use viewer pan/fit and download full-size SVG. - Pathway grouped bars can be long; viewer supports vertical scroll for full content. - UI detail caps can differ from full exported row counts by design. ## Graph-by-Graph Reference ### 1) Main Network (`network_main`) - Answers: Which molecules are linked in the active query context? - Inputs: Network nodes/edges from `/api/pathway` and detail expansion events. - Encoding: - Node color/shape by molecule type (`miRNA`, `Gene`, `Protein`) - Edge direction reflects regulatory direction where available - Interaction controls: pan/fit in analysis and popup viewer; PNG/SVG export. - Pitfalls: Visual centrality is layout-dependent and is not a statistical centrality score. - Related exports: `01_network/graph_nodes_full.csv`, `01_network/graph_edges_full.csv`. ### 1b) Right-Panel PPI Tree (`ppi_tree_panel`) - Answers: Which one-hop protein partners are most relevant for the selected protein? - Inputs: `/api/ppi_network` rows for selected protein, plus active UI threshold/search/cap. - Encoding: - Row label: partner protein - Badge: `combined_score` - Mini badges: selected evidence channels (`experimental`, `database`, `textmining`) when non-zero - Controls: - collapsed by default - in-panel `Show On Graph` ON/OFF toggle with status badge - `Clear Overlay` action to remove projected PPI-only nodes/edges - `Show 20` reset and `Show +10` progressive reveal - threshold chips (`700`, `800`, `900`) - search filter for partner labels - click row to focus node in graph; if missing and overlay is ON, EVd3x projects and focuses it - Pitfalls: - one-hop only by design for readability - view may be capped while full table remains exportable - overlay projection is intentionally capped for readability and performance - Related exports: - `09_ppi_analysis/ppi_tree_full.csv` - `09_ppi_analysis/ppi_tree_view.csv` ### 2) Node EV Evidence Sources (`node_ev_sources`) - Answers: Which EV evidence databases support the selected node? - Inputs: Node `ev_evidence` rows, grouped by `Source_Database`. - Encoding: - X-axis: source database - Y-axis: evidence row count - Controls: popup viewer and PNG/SVG download. - Pitfalls: evidence row count is not equivalent to unique study count without publication-level collapse. - Related exports: `02_node_details/node_ev_evidence.csv`. ### 3) Target Confidence Distribution (`node_target_confidence`) - Answers: How strong are miRNA-target confidences for the selected miRNA? - Inputs: node-level `graph_elements` with target edges and `Confidence_Score`. - Encoding: - X-axis: target rank/bin (implementation dependent) - Y-axis: confidence score - Controls: popup viewer and PNG/SVG download. - Pitfalls: scores are integrated confidence values; compare within the same query context. - Related exports: `01_network/graph_edges_full.csv`, `07_pathway_analysis/pathways_full.csv`. ### 4) Cell Communication Heatmap (`cellcomm_heatmap`) - Answers: Which source-target cell pairs show strongest communication support? - Inputs: `/api/cell_communication_v2` heatmap matrix and selected cell sets. - Encoding: - X-axis: target cell - Y-axis: source cell - Color: communication score - Controls: click cell to open detailed interaction table; popup export. - Pitfalls: displayed detail interactions may be capped; inspect manifest + full export for totals. - Related exports: - `05_cell_communication/heatmap_matrix_full.csv` - `05_cell_communication/interactions_full.csv` - `05_cell_communication/interactions_selected_pair_view.csv` ### 5) L-R Bipartite (`lr_bipartite`) - Answers: How do query genes map onto ligand/receptor partner genes? - Inputs: `/api/ligand_receptor_analysis` result tables. - Encoding: - Left partition: query molecules with LR role - Right partition: partner molecules - Link: ligand->receptor relation - Controls: popup viewer, drag/pan, fit/reset, PNG/SVG export. - Pitfalls: dense systems can exceed label readability; increase focus using filters and exports. - Related exports: `06_lr_analysis/query_as_ligands.csv`, `06_lr_analysis/query_as_receptors.csv`. ### 6) Pathway Grouped Bars (`pathway_bars`) - Answers: Which pathways are most enriched within each pathway category? - Inputs: filtered pathway enrichment results from `/api/pathway_enrichment`. - Encoding: - Group: pathway category - Bars: pathways within category - Value: enrichment signal (rank/statistic as rendered) - Controls: full-content viewer with vertical scrolling, PNG/SVG export. - Pitfalls: this chart follows active filters (source and p-value); compare only within same filter state. - Related exports: `07_pathway_analysis/pathways_filtered_view.csv`, `07_pathway_analysis/pathways_full.csv`. ### 7) Disease Treemap (`disease_treemap`) - Answers: Which disease groups dominate evidence for the current query/system? - Inputs: grouped disease associations keyed by `Disease_ID`. - Encoding: - Rectangle size: relative evidence count/weight - Rectangle color: disease category group - Controls: popup viewer and image export. - Pitfalls: category names are display-normalized; canonical key remains `Disease_ID`. - Related exports: - `08_disease_analysis/disease_bubble_summary_full.csv` - `08_disease_analysis/disease_associations_full.csv` - `08_disease_analysis/disease_associations_filtered_view.csv` ### 8) Cell Specificity Lollipop (`cellspec_lollipop`) - Answers: Which cell types rank highest for the query set? - Inputs: `/api/cell_specificity` ranked table. - Encoding: - X-axis: specificity score - Y-axis: cell type (top-ranked) - Controls: viewer export and table cross-check. - Pitfalls: rank stability depends on query composition and source coverage. - Related exports: `04_cell_specificity/cell_specificity_ranking_full.csv`. ### 9) Cell Specificity by System (`cellspec_system`) - Answers: Which organ/system groups contribute most to specificity signal? - Inputs: aggregated cell specificity output by system labels. - Encoding: - X-axis: system - Y-axis: aggregate score/coverage - Controls: viewer export. - Pitfalls: systems are curated grouping labels, not independent biological ontologies. - Related exports: `04_cell_specificity/cell_specificity_selected_cells.csv`. ## Figure Mode (Phase 6) EVd3x includes a **Figure Mode** toggle in the top header bar (camera icon, labelled "Figure"). When Figure Mode is ON: - A metadata bar appears below the header showing current query, mode, direction, Top-N, active lens filters, selected cells, and communication scoring context. - The **Copy Methods Snippet** button generates a structured text block with all active filter settings, database credits, and scoring parameters, ready to paste into figure legends or methods sections. - Non-essential chrome is reduced for cleaner figure captures. - Graph legend and export controls are more prominent. Figure Mode does **not** change any scientific calculations. It is a presentation/export overlay only. ## Graph Viewer Export Notes (Phase 6) When opening any graph in the viewer, a context note is shown above the plot with: - Current query and mode - Active p-value cutoff (for pathway views) - Truncation warning if cell communication view was capped **Export format guidance:** - **SVG**: recommended for publication because it is vector, scalable, and editable in Illustrator or Inkscape. - **PNG**: quick preview and slide use; resolution depends on screen density. Always use **Fit to View** before exporting to ensure no content is clipped. ## Export bundle contents (Phase 6) The full analysis ZIP now includes: - `00_metadata/session_metadata.csv`: query, mode, direction, Top-N, timestamp, app version. - `00_metadata/figure_settings.csv`: snapshot of all active lens filters at export time, including pathway p-value, sources, link cap, keyword, disease filters, selected cells, communication scoring, and PPI settings. - `99_meta/figure_settings` in manifest: same settings in manifest row format. - `99_meta/processing_notes`: backend processing details such as canonicalization, grouping keys, and motif rule version. ## Canonical Demo Workflow (Manuscript Figures) Use these query sets for reproducible manuscript figures: **Figure 1: Alzheimer natural-language entry:** - Query: `early onset Alzheimer's disease with behavioral disturbance` - Mode: disease-first natural-language route - Visible first-pass state: PSEN1 mRNA, PSEN1 protein, hsa-miR-4496, hsa-miR-107, hsa-miR-4677-3p - Note: the 25 molecule value is an interactive processing guard, not a biological target. **Figure 2: PSEN1 selected-node evidence:** - Selected node: `PSEN1 protein` - Surfaces: main graph, node details, EV evidence, EV-TRACK-linked publication context, localization, PPI neighborhood - Note: EV evidence rows are reported detections, not functional transfer claims. **Figure 3: Alzheimer disease convergence:** - Query state: PSEN1-centered Alzheimer analysis state - Surfaces: disease categories, ranked disease groups, selected disease evidence rows - Note: report disease rows as association evidence, not diagnosis. **Figure 4: Pathway analyses:** - Query state: direct seeds plus expanded target and network-neighbor genes - Surfaces: pathway category map and pathway detail table - Filters: hypergeometric `p <= 0.05`, BH-FDR `q <= 0.10`; record source resources and link cap. **Figure 5: Cell specificity + communication + LR:** - Query state: Alzheimer analysis state - Surfaces: cell specificity ranking, nervous-system focus, oligodendrocyte to astrocyte communication, PSEN1-NOTCH examples - Note: record selected cells and communication relevance formula. **Figure 6: LR, miRNA-target, and PPI bridges:** - Selected interaction: PSEN1-centered LR/PPI bridge state, including PSEN1 to CTNNB1 - Surfaces: LR arcs, direct miRNA-target overlap, shared target bridges, high-confidence STRING neighborhood - Note: record miRNA bridge class, support count, mean/max confidence, and PPI threshold `combined_score >= 700`. ## Processing transparency checks Before interpretation or publication: 1. Confirm if view is truncated (`total_rows` vs `returned_rows`, manifest notes). 2. Use Figure Mode → Copy Methods Snippet to capture all active filter settings. 3. Check `00_metadata/figure_settings.csv` to verify what was active at export time. 4. Use CSV files for quantitative analysis; treat charts as visual summaries. 5. Include source provenance fields (`Source_Database`, `Pathway_Source`, `Source`) in downstream analysis. 6. For right-panel PPI interpretation, report active threshold and view cap from export manifest notes. 7. SVG exports are the recommended archival format for all publication figures.