Open 📂 Data → Load Demo . A brief modal describes the dataset; loading runs in the background. Close the modal when you are ready. Five tissue-architecture channels appear in 📋 Channels — H&E Image, Nuclei, Fibrillar ECM, Soft Tissue, and Entropy. Click each to inspect them. Use mouse-wheel to zoom, click-drag to pan, ⊙ to reset.

Open ⬡ Acquire . Type ERBB2 and select it from the dropdown — it closes automatically on selection. Choose red as the colour. Leave Auto on. Expand 🧹 Leakage correction — the slider is at α = 0.10 by default. Click ⚡ Auto-detect α to let NaVis sweep α values and find the architecture-coupling elbow for this gene (typically 0.05–0.15 on FFPE Visium). Then click ⬡ Acquire Channel Image . The map appears in Channels and is selected automatically.

The newly acquired ERBB2 channel still carries some halo from low-confidence spots — the AD-EBIDW interpolation extends each spot's signal a few hundred microns beyond its physical footprint, and the lowest-intensity tail of that extension is where most spurious-looking signal lives. Open 🎚 Image Controls to clean it up live.

• Noise gate — pixels below this intensity are forced to zero. Default 0.30; raise to 0.40–0.50 to drop diffuse background and keep only the regions where the gene was confidently expressed. The channel image redraws live as you drag the slider.
• Contrast boost — applies an unsharp-mask-style boost to the surviving signal. Default 0 (off); raise to 0.3–0.6 after gating to recover crispness of the regions you kept.

Three buttons under Tools expose secondary readouts:

• 📈 Spatial Autocorrelation — computes Moran's-style autocorrelation of the active channel at multiple lag distances and plots the decay profile. The shape of the curve characterises the channel's spatial scale: a sharp drop signals fine-grained patterning, a slow decay signals broad gradients. Useful for comparing how 'clumpy' vs 'smooth' two channels are.
• 🎨 Show Legend — displays the colour-mapping legend for the active channel (the gene name, the colour assigned, the dynamic range). The legend is also embedded in any saved channel ZIP.
• 📋 Image Parameters — opens a modal listing every parameter used to build the active channel: the gene or signature name, leakage-correction α, noise gate, contrast boost, AD-EBIDW σ and iteration count, dropout sensitivity τ (for signatures), and any leakage-correction status. This is the channel's audit trail — paste it into a notebook for reproducibility.

Image Controls operate on the rendered image, not on the raw expression data — Insights and Find-Similar-Genes always re-read the underlying channel. Adjustments here are for visualisation; the analytical readouts are unaffected. Each channel keeps its own slider settings, so switching channels restores their independent values.

Open 🔭 Insights → click ▶ Run Spatial Analysis . Observed results on this dataset (ERBB2, auto α = 0.15, auto mode):

• Compartment Enrichment — near 1× across all three compartments. An epithelial marker like ERBB2 occupies cellular regions generally; the H&E-derived priors (nuclei density, fibrous ECM, soft tissue) are architectural signals that do not separate epithelial identity specifically, so 1× is the honest null. Strong enrichment requires a signature specific to a compartment- bound phenotype.
• Expression Boundaries — gradient map shows sharp ductal perimeters. Boundary alignment with tissue edges is moderate (~0.2–0.3 Pearson r for all three priors), indicating ductal boundaries align with general tissue-architecture transitions rather than with any single compartment edge specifically.
• Spatial Cross-correlation — peaks around r ≈ 0.5 at lag 0 µm for Nuclei (ERBB2+ cells contain nuclei), decays monotonically, crosses zero around 400 µm, and goes negative beyond that. The 400 µm zero-crossing reflects this section's typical duct-to-duct spacing; at longer distances ducts are anti-correlated with each other because they are separated by stroma.
• Expression Islands — 102 islands above the 75th-percentile threshold, median area 16 µm², mean area 5 783 µm², largest island 586 000 µm², median circularity 0.79. The huge mean vs median gap signals a heavily skewed size distribution: many small intra-ductal hotspots plus a few very large connected components that span entire ducts or groups of ducts merged across the threshold.
• Histology Decoupling — 63% (MODERATE). ERBB2's spatial pattern is partly explained by the H&E priors but also carries genuine transcriptional information beyond histology — expected for a gene whose level varies between histologically similar ducts.

With ERBB2 still selected, click 🔎 Find Similar Genes . Correlation runs over all spots (including zeros), so a gene expressed everywhere scores near zero — only genes with matching spatial ON/OFF patterns rank high.

Observed top 10 for ERBB2 (Pearson r, spot-level):

Nine of ten hits are canonical luminal/ductal epithelial markers ( AZGP1 , CD24 , MUCL1 , KRT7 , KRT8 , FOXA1 , etc.) — this is strong biological validation: the spatial correlation recovered the HER2+ ductal-epithelial compartment from scratch without any prior knowledge. Click any gene name in the results to load it into ⬡ Acquire and visualise its spatial pattern.

Acquire ACTA2 in green — a microenvironmental marker that highlights α-smooth-muscle-actin-expressing cells surrounding the ductal compartment (cancer-associated fibroblasts, myofibroblasts and myoepithelial cells all contribute). The resulting pattern is typically ductal-adjacent rather than a sharp perimeter ring — use it to visualise the tumour–microenvironment interface. Open ⊕ Merge & Interface , set Channel A to ERBB2 and Channel B to ACTA2, optionally adjust the reach (μm) slider, then click Compute . Three new channels appear in 📋 Channels : the composite (blended A+B colors), the agreement map (bright where A and B overlap), and the new interface map (bright at the boundary between A's and B's territories, in each channel's color). The Last computation section of the Merge panel reports interface area, A-vs-B asymmetry, and mean interface strength. Open 🎚 Image Controls to adjust per-layer noise gate and contrast independently.

Open 🔆 H&E Overlay . Select ERBB2 from the dropdown. Move the opacity and brightness sliders — the preview updates live in the panel. When satisfied, click 🔆 Acquire this Overlay to commit it as a permanent channel that can be used in further merges and analysis.

Click 🔬 Analyze in the toolbar to enter Analyze mode. The Analyze panel hosts three distinct interactions, each tied to a different mouse action. Press Escape to exit at any time.

After running Insights, click 💾 Save all as ZIP to download all five analysis tables (compartment enrichment CSV, boundary alignment CSV, cross-correlation CSV, island morphology CSV, summary TXT with decoupling score) plus the boundary gradient PNG. Individual sections also have per-section 💾 download links. Zoom any panel with 🔍 for a larger, print-quality view of each result.