跳到主要内容

热力图

热力图把一张二维网格画出来,每格的颜色编码一个数值。数值会归一到数据自身的范围、再过一遍色图,右侧画一条 色条。

WITH d AS (
SELECT gene, [Sample_01, Sample_02, Sample_03, Sample_04, Sample_05, Sample_06,
Sample_07, Sample_08, Sample_09, Sample_10, Sample_11, Sample_12] AS row_vals
FROM read_csv_auto('{{DFK_BASE_URL}}data/heatmap.tsv')
)
SELECT kuva_render(to_json({
'title': 'Expression heatmap',
'series': [{
'type': 'heatmap',
'data': (SELECT list(row_vals ORDER BY gene) FROM d),
'row_labels': (SELECT list(gene ORDER BY gene) FROM d),
'col_labels': ['Sample_01', 'Sample_02', 'Sample_03', 'Sample_04', 'Sample_05', 'Sample_06',
'Sample_07', 'Sample_08', 'Sample_09', 'Sample_10', 'Sample_11', 'Sample_12'],
'color_map': 'viridis'
}]
})) AS chart;

data 是行优先的:data[row][col],外层列表从上往下、内层从左往右。一张宽表就是「一行一个列表」—— 对数值列做一次 list([…]),行序由你排。

轴标签​

row_labels 与 col_labels 给格子命名。row_labels 是从下往上的(y 轴的惯例),所以第一个标签落在图的 底部;col_labels 从左往右。

WITH d AS (
SELECT gene, [Sample_01, Sample_02, Sample_03, Sample_04, Sample_05, Sample_06,
Sample_07, Sample_08, Sample_09, Sample_10, Sample_11, Sample_12] AS row_vals
FROM read_csv_auto('{{DFK_BASE_URL}}data/heatmap.tsv')
ORDER BY gene DESC
)
SELECT kuva_render(to_json({
'title': 'Labelled rows',
'series': [{
'type': 'heatmap',
'data': (SELECT list(row_vals ORDER BY gene DESC) FROM d),
'row_labels': (SELECT list(gene ORDER BY gene DESC) FROM d),
'col_labels': cols,
'color_map': 'inferno'
}]
})) AS chart
FROM (
SELECT ['Sample_01', 'Sample_02', 'Sample_03', 'Sample_04', 'Sample_05', 'Sample_06',
'Sample_07', 'Sample_08', 'Sample_09', 'Sample_10', 'Sample_11', 'Sample_12'] AS cols
);

标签列表要和矩阵同一个顺序 —— 两者对不上,你会得到一张画得完美、标得全错的图。

写上数值​

show_values 把每格的原始数值写在格子里。小网格值得开,大网格根本读不了。

SELECT kuva_render(to_json({
'x_axis': {'name': 'sample'},
'y_axis': {'name': 'gene'},
'series': [{
'type': 'heatmap',
'data': [
[10.0, 20.0, 30.0, 15.0],
[45.0, 55.0, 25.0, 60.0],
[70.0, 35.0, 80.0, 40.0],
[50.0, 90.0, 65.0, 20.0]
],
'row_labels': ['GeneA', 'GeneB', 'GeneC', 'GeneD'],
'col_labels': ['Ctrl', 'T1', 'T2', 'T3'],
'show_values': true,
'color_map': 'grayscale'
}]
})) AS chart;

色图​

color_map色阶说明
"viridis"蓝 → 绿 → 黄感知均匀、对色盲友好。(默认)
"inferno"黑 → 紫 → 黄对比强;灰度打印也还读得出来
"grayscale"黑 → 白干净的出版风格

色图里的名字都能用;legend 给色条加标题。

WITH d AS (
SELECT gene, [Sample_01, Sample_02, Sample_03, Sample_04, Sample_05, Sample_06,
Sample_07, Sample_08, Sample_09, Sample_10, Sample_11, Sample_12] AS row_vals
FROM read_csv_auto('{{DFK_BASE_URL}}data/heatmap.tsv')
)
SELECT kuva_render(to_json({
'title': 'With a colour bar title',
'series': [{
'type': 'heatmap',
'data': (SELECT list(row_vals ORDER BY gene) FROM d),
'row_labels': (SELECT list(gene ORDER BY gene) FROM d),
'color_map': 'grayscale',
'legend': 'z-score'
}]
})) AS chart;

自定义坐标范围​

默认列映射到 [0.5, 列数 + 0.5]、行映射到 [0.5, 行数 + 0.5],这样整数刻度正好落在格心。当这张网格代表一个 物理区域时,用 x_range / y_range 把真实坐标放到轴上。

WITH g AS (
SELECT i AS row, j AS col,
((j + 0.5) * 20.0 / 40.0) - 10.0 AS x,
4.0 - ((i + 0.5) * 8.0 / 16.0) AS y
FROM (SELECT unnest(range(0, 16)) AS i), (SELECT unnest(range(0, 40)) AS j)
),
rows AS (
SELECT row, list(exp(-(x * x / 16.0 + y * y / 4.0) / 2) ORDER BY col) AS row_vals
FROM g
GROUP BY row
)
SELECT kuva_render(to_json({
'title': 'Scalar field',
'x_axis': {'name': 'x (m)'},
'y_axis': {'name': 'y (m)'},
'series': [{
'type': 'heatmap',
'data': (SELECT list(row_vals ORDER BY row) FROM rows),
'color_map': 'inferno',
'x_range': [-10, 10],
'y_range': [-4, 4]
}]
})) AS chart;

两侧也可以只给一边 —— 只钉 x 而让 y 留在整数刻度上,或者反过来。

格子大小​

cell_size 是每格矩形占自己槽位的比例。默认 0.99 留一道发丝缝,格子的边界看得见;1.0 是严丝合缝、没有网格线 —— 大矩阵上就该用它,否则那些缝会变成一片干扰性的花纹。

WITH d AS (
SELECT gene, [Sample_01, Sample_02, Sample_03, Sample_04, Sample_05, Sample_06,
Sample_07, Sample_08, Sample_09, Sample_10, Sample_11, Sample_12] AS row_vals
FROM read_csv_auto('{{DFK_BASE_URL}}data/heatmap.tsv')
)
SELECT kuva_render(to_json({
'title': 'Flush cells',
'series': [{
'type': 'heatmap',
'data': (SELECT list(row_vals ORDER BY gene) FROM d),
'color_map': 'magma',
'cell_size': 1.0
}]
})) AS chart;

行序​

热力图好不好用,全看行序。这里没有隐藏的排序:让矩阵与标签按同一个顺序排,画出来就是这个顺序。按行的均值排序是 最经典的一招 —— 它把矩阵变成一道一眼就能读的渐变。

WITH d AS (
SELECT gene,
[Sample_01, Sample_02, Sample_03, Sample_04, Sample_05, Sample_06,
Sample_07, Sample_08, Sample_09, Sample_10, Sample_11, Sample_12] AS row_vals,
(Sample_01 + Sample_02 + Sample_03 + Sample_04 + Sample_05 + Sample_06
+ Sample_07 + Sample_08 + Sample_09 + Sample_10 + Sample_11 + Sample_12) / 12 AS row_mean
FROM read_csv_auto('{{DFK_BASE_URL}}data/heatmap.tsv')
)
SELECT kuva_render(to_json({
'title': 'Rows sorted by mean',
'series': [{
'type': 'heatmap',
'data': (SELECT list(row_vals ORDER BY row_mean) FROM d),
'row_labels': (SELECT list(gene ORDER BY row_mean) FROM d),
'color_map': 'viridis',
'legend': 'z-score'
}]
})) AS chart;

字段​

字段类型设置什么
datanumber[][]必填。 矩阵,行优先:data[row][col]。每行必须等长。
row_labelsstring[]行标签(y 轴,从下往上)。
col_labelsstring[]列标签(x 轴,从左往右)。
color_mapstring色图。
show_valuesboolean把每格的数值写在格子里。
x_range / y_range[number, number]给坐标轴一套自定义坐标(默认 [0.5, n + 0.5])。
cell_sizenumber格子填充比例,内部夹到 [0.5, 1](默认 0.99)。
legendstring色条的标题。
tooltipsboolean悬停提示。
tooltip_labelsstring[]每格一条提示文字,按行优先顺序。

热力图没有 color 字段 —— 颜色本身就是编码,所以选颜色只能靠 color_map。

说明​

  • data 的每一行必须等长,且矩阵不能为空。
  • 给了 row_labels / col_labels 就要与行数 / 列数一致 —— 而且顺序必须与 data 对齐,因为不会替你重排。
  • cell_size 会被夹到 [0.5, 1]。
  • 数值是按矩阵自己的最小最大值归一的,所以两张取值范围不同的热力图并不可比,除非你用 x_range / y_range 把坐标钉住 —— 色标跟的是数据,不是坐标范围。

另见​