主题
视网膜单细胞
步骤 6-4 · 只标注 · 图 3 张 · 发表版表格 0 个 （流程见 v5 定稿）
3,177,310 个细胞核的视网膜图谱：候选基因在哪些细胞类型里表达。⚠️ 该图谱没有内皮细胞与周细胞，对血管病是硬伤。
图
Figure F6-4b-a. Expression of the 11 candidate genes across the 31 retinal cell types of the Human Retina Cell Atlas
完整图注（英文，与投稿版一致）
Figure F6-4b-a. Expression of the 11 candidate genes across the 31 retinal cell types of the Human Retina Cell Atlas.
Panel. One dot per gene and cell type (341 dots, no combination missing). Dot area is the fraction of nuclei of that cell type with a non-zero count for that gene; dot colour is the mean normalised expression across all nuclei of that cell type, including the non-expressing ones. Cell types are ordered by the number of nuclei in the atlas, most abundant at the top; genes are ordered by their mean expressing fraction across cell types, broadest on the left.
Sampling. Unlike the two UMAP figures, this panel is computed over all 3,177,310 nuclei and is not downsampled; every nucleus of every cell type contributes to both the expressing fraction and the mean. Records whose stored value is exactly zero are counted as non-expressing.
Atlas. Human Retina Cell Atlas, single-nucleus RNA transcriptome (hrca_snrna_allcells.h5ad; 3,177,310 nuclei, 35,475 genes, 31 annotated cell types). Yan et al., Single-cell atlas of the transcriptome and chromatin accessibility in the human retina, Nature Genetics 2025 (PMID 41578023). Cell-type annotation is taken from the atlas as published and was not recomputed here. The expression matrix is the atlas normalised matrix, not raw counts.
⚠ Known gap. This atlas contains no endothelial cells and no pericytes (no_endothelial_no_pericyte_in_this_atlas). Diabetic retinopathy, maculopathy and nephropathy are microvascular diseases, so the cell types most likely to carry the relevant signal are absent from this panel. A negative result here therefore cannot be read as absence of expression in the retina; it means absence in the cell types this atlas samples. This limitation cannot be repaired by a different atlas of the same kind and is stated in the Limitations section.
What this figure is and is not. It is step-6 descriptive annotation: it locates where the candidate genes are transcribed. It is not a causal analysis. No Mendelian randomization, no colocalisation and no candidate was promoted, demoted or vetoed on the basis of this figure. No significance threshold applies to this figure; nothing here is tested.
Data sources and sample sizes. Exposure: UKB-PPP plasma proteome (Olink Explore 3072), 34,557 European participants, 1,954 proteins with a cis-pQTL.
Figure F6-4b-b. Uniform manifold approximation and projection (UMAP) of the Human Retina Cell Atlas, with the 31 annotated cell types labelled
完整图注（英文，与投稿版一致）
Figure F6-4b-b. Uniform manifold approximation and projection (UMAP) of the Human Retina Cell Atlas, with the 31 annotated cell types labelled.
Panel. The UMAP coordinates are those published with the atlas and were not recomputed. Each point is one nucleus, coloured by the atlas major class, of which there are 10. Colour therefore encodes the coarse grouping; the finer identity of all 31 cell types is carried by the labels, each placed at the median UMAP position of its cell type. Colour is not used for the 31 cell types themselves because no categorical palette is reliably distinguishable at that number of levels. The number in each legend entry is the count of nuclei of that major class in the full atlas, not in the plotted sample.
Major classes and their member cell types. Rod: retinal rod cell. BC: diffuse bipolar 1 cell, diffuse bipolar 2 cell, diffuse bipolar 3a cell, diffuse bipolar 3b cell, diffuse bipolar 4 cell, diffuse bipolar 6 cell, flat midget bipolar cell, giant bipolar cell, invaginating midget bipolar cell, OFFx cell, ON-blue cone bipolar cell, retinal bipolar neuron, rod bipolar cell. AC: amacrine cell, GABAergic amacrine cell, glycinergic amacrine cell, starburst amacrine cell. RGC: OFF midget ganglion cell, OFF parasol ganglion cell, ON midget ganglion cell, ON parasol ganglion cell, retinal ganglion cell. MG: Mueller cell. Cone: retinal cone cell, S cone cell. HC: H1 horizontal cell, H2 horizontal cell. Astrocyte: astrocyte. Microglia: microglial cell. RPE: retinal pigment epithelial cell.
Colour palette, and how it was chosen. Six colours are taken from the colour-blind-safe eight-colour palette used throughout this paper and four were added, because the atlas has 10 major classes. Two of the eight were deliberately left out: black, which would be given to the most abundant class and would dominate the panel, and the pale yellow, whose lightness differs from the white background by only 10.9 units on the CIE L* axis against at least 22.8 for every colour that was kept. The four added colours were selected by searching a hue-chroma-lightness grid under a lightness constraint, so that the ten colours span a range of only 32 lightness units rather than the 71 that an unconstrained search produced. Separability was then verified rather than assumed: the smallest pairwise CIE Lab distance among the ten, evaluated under normal vision and under simulated deuteranopia, protanopia and tritanopia, is 16.2, equal to the smallest such distance within the paper's own eight-colour palette. Colours are assigned so that the most abundant class receives the visually lightest colour and the rarest the heaviest: abundance is already conveyed by the area a class occupies, so colour weight is given to the classes that would otherwise be overlooked.
Sampling. The background layer is a fixed-seed random sample of 300,000 of the 3,177,310 nuclei (seed 1), taken because plotting every nucleus is illegible at print size and produces a file that exceeds the journal size limit. The sample is used for the point cloud and for the cell-type median positions only.
Atlas. Human Retina Cell Atlas, single-nucleus RNA transcriptome (hrca_snrna_allcells.h5ad; 3,177,310 nuclei, 35,475 genes, 31 annotated cell types). Yan et al., Single-cell atlas of the transcriptome and chromatin accessibility in the human retina, Nature Genetics 2025 (PMID 41578023). Cell-type annotation is taken from the atlas as published and was not recomputed here. The expression matrix is the atlas normalised matrix, not raw counts.
⚠ Known gap. This atlas contains no endothelial cells and no pericytes (no_endothelial_no_pericyte_in_this_atlas). Diabetic retinopathy, maculopathy and nephropathy are microvascular diseases, so the cell types most likely to carry the relevant signal are absent from this panel. A negative result here therefore cannot be read as absence of expression in the retina; it means absence in the cell types this atlas samples. This limitation cannot be repaired by a different atlas of the same kind and is stated in the Limitations section.
What this figure is and is not. It is step-6 descriptive annotation: it locates where the candidate genes are transcribed. It is not a causal analysis. No Mendelian randomization, no colocalisation and no candidate was promoted, demoted or vetoed on the basis of this figure. No significance threshold applies to this figure; nothing here is tested.
Data sources and sample sizes. Exposure: UKB-PPP plasma proteome (Olink Explore 3072), 34,557 European participants, 1,954 proteins with a cis-pQTL.
Figure F6-4b-c. Single-nucleus expression of each of the 11 candidate genes on the retinal UMAP
完整图注（英文，与投稿版一致）
Figure F6-4b-c. Single-nucleus expression of each of the 11 candidate genes on the retinal UMAP.
Panels. One panel per gene, on the same UMAP coordinates as the previous figure. Grey points are the 300,000-nucleus background sample. Coloured points are nuclei with a non-zero normalised count for that gene, shaded by the value. PAM 1,094,067; IFNAR1 608,960; NUDT5 613,934; APOE 514,261; NOTCH2 452,412; WARS 400,370; LACTB2 212,663; ERMAP 139,353; GALNT3 102,528; ACRBP 20,387; SIGLEC5 12,551 expressing nuclei respectively. Panels follow the same gene order as the dot-plot figure, so the two can be read column by column. Both axes are the published UMAP coordinates and are drawn on a common fixed aspect ratio; the axes carry no units and are therefore not labelled.
Colour scale, and why it is clipped. Expression is shown on a linear scale that saturates at 2.32, the 99th percentile of all non-zero values; the 1% of nuclei above it are drawn at the top colour and the legend is marked accordingly. Without this the scale would run to 4.88 while the median non-zero value is 0.75, so more than nine in ten points would fall in the lowest quarter of the colour range and all eleven panels would look identical. This is a display choice; no value was altered.
★ Sampling asymmetry, deliberate. The grey background is downsampled (see below) but the expressing nuclei are not: all 4,171,486 records are drawn. Downsampling the expressing layer would render a lowly expressed gene indistinguishable from an unexpressed one, which is exactly the distinction this figure exists to show. Absolute point density is therefore not comparable between the grey and the coloured layer.
Sampling. The background layer is a fixed-seed random sample of 300,000 of the 3,177,310 nuclei (seed 1), taken because plotting every nucleus is illegible at print size and produces a file that exceeds the journal size limit. The sample is used for the point cloud and for the cell-type median positions only.
Atlas. Human Retina Cell Atlas, single-nucleus RNA transcriptome (hrca_snrna_allcells.h5ad; 3,177,310 nuclei, 35,475 genes, 31 annotated cell types). Yan et al., Single-cell atlas of the transcriptome and chromatin accessibility in the human retina, Nature Genetics 2025 (PMID 41578023). Cell-type annotation is taken from the atlas as published and was not recomputed here. The expression matrix is the atlas normalised matrix, not raw counts.
⚠ Known gap. This atlas contains no endothelial cells and no pericytes (no_endothelial_no_pericyte_in_this_atlas). Diabetic retinopathy, maculopathy and nephropathy are microvascular diseases, so the cell types most likely to carry the relevant signal are absent from this panel. A negative result here therefore cannot be read as absence of expression in the retina; it means absence in the cell types this atlas samples. This limitation cannot be repaired by a different atlas of the same kind and is stated in the Limitations section.
What this figure is and is not. It is step-6 descriptive annotation: it locates where the candidate genes are transcribed. It is not a causal analysis. No Mendelian randomization, no colocalisation and no candidate was promoted, demoted or vetoed on the basis of this figure. No significance threshold applies to this figure; nothing here is tested.
Data sources and sample sizes. Exposure: UKB-PPP plasma proteome (Olink Explore 3072), 34,557 European participants, 1,954 proteins with a cis-pQTL.
发表版表格
本页的表格与 [靶点 pheWAS 与成药性](/dm-complications/29-靶点 pheWAS 与成药性) 共用（同属 09_descriptive_annotation）。
本页图为网页版 PNG（最宽 1600 px）。投稿用的矢量 PDF 与 600 dpi TIFF 体积较大，留在仓库
results/singlecell/figures/，不随文档站分发。