AI复现顶刊特色同款：山脊图+热图的组合

TL;DR - A WeChat bioinformatics tutorial showing how to use an AI assistant (with a custom "xinye-plot" plotting skill) to reproduce Figure 2d of a 2026 Nature Cell Biology primate embryo spatial-transcriptomics paper — a ridgeline + heatmap combo of HOX expression along the anterior–posterior axis. It matters as a concrete example of LLM-assisted scientific figure reproduction from published source data.

• Source figure comes from a CS9 macaque embryo Stereo-seq dataset (10 near-midline sagittal sections, bin50 ≈ 25 μm, 35,568 QC-passed spots); after 3D reconstruction, bins were ranked by Euclidean distance from an anterior reference point and binned into 100 A–P layers, then averaged per HOX gene.
• Plotted data is the paper's own source file (..._source_data_fig2.xlsx, sheet "Panel d": 31 HOX genes × AP_1–AP_100); raw sequencing is deposited at CNGBdb under CNP0007017 — not GSE193007.
• Biological reading: anterior HOX genes (HOXA1, HOXB1, HOXB2) show broad A–P coverage, while central/posterior genes (HOXA7/A9, HOXB9, HOXC10, HOXD13) peak progressively posteriorly and more narrowly, supporting progressive axial regionalization; HOXA10/A11/A13 also appear in the allantois.
• Reproduction uses R (readxl, ggplot2, ggridges, patchwork, cowplot, scales): per-gene min–max normalization for ridge height only, smooth.spline (spar 0.45) plus a cubic Hermite tail extension for display, geom_ridgeline with a manual palette, and a row z-scored geom_tile heatmap with a diverging blue–white–red scale (limits ±4, squished).