Issue in circular heatmap, the rows value are getting interchanged.

Question

[Yukta]

I have a vst file for which I am generating a circular heatmap. The heatmap is being generated but in the image the values of "A" is being changed with "E".

Like for example; some of my values in "A" are positive like fat.1, fat.2 etc. but in the heatmap they are depicted as negative.

I am sharing my R script for which I have generated , but I am not sure why the column values are getting interchanged.

R script:

# Load required libraries
library(circlize)
library(ComplexHeatmap)
library(dendextend)
library(grid)

# ------------------- Data Preprocessing -------------------
#read the scaled data
vst_scaled <- read.csv("ALL/L2FC1/VST_scaled_for_heatmap_updated.csv", row.names = 1)
mat_t <- as.matrix(vst_scaled)

# ------------------- Clustering and Ordering -------------------
# Cluster genes (rows) and reorder
dend <- as.dendrogram(hclust(dist(mat_t)))
gene_order <- order.dendrogram(dend)
mat_t <- mat_t[gene_order, ]

# Color mapping function
col_fun <- colorRamp2(c(-3, 0, 3), c("blue", "white", "red"))

# Prepare legends
# Prepare legends with bold, larger fonts
lgd_links <- Legend(
  at = c(-3, -1.5, 0, 1.5, 3),
  col_fun = col_fun,
  title = "z-score",
  direction = "horizontal",
  title_position = "topleft",
  labels_gp = gpar(fontsize = 16, fontface = "bold"),
  title_gp = gpar(fontsize = 18, fontface = "bold")
)

lgd_annotation <- Legend(
  labels = c("High", "Low"),
  title = "Expression",
  legend_gp = gpar(fill = c("red", "blue")),
  direction = "horizontal",
  title_position = "topleft",
  labels_gp = gpar(fontsize = 16, fontface = "bold"),
  title_gp = gpar(fontsize = 18, fontface = "bold")
)

# Add a custom sample legend for treatments
lgd_samples <- Legend(
  labels = c("A = Control", "B = NN-Extract", "C = NN@AuNPs", "D = Au@NPs", "E = Orlistat"),
  title = "Treatments",
  direction = "horizontal",
  title_position = "topleft",
  legend_gp = gpar(fill = NA),
  labels_gp = gpar(fontsize = 16, fontface = "bold"),
  title_gp = gpar(fontsize = 18, fontface = "bold")
)

# Combine all legends vertically
lgd_combined <- packLegend(lgd_links, lgd_annotation, lgd_samples, direction = "vertical")

# Save to high-resolution PNG (600 DPI)
png("Circos_Heatmap_VST_scaled.png", width = 16, height = 10, units = "in", res = 600)

# Set up layout: 75% plot (left), 25% legend (right)
pushViewport(viewport(layout = grid.layout(nrow = 1, ncol = 2, widths = unit(c(3, 1), "null"))))

# Circos heatmap (left)
pushViewport(viewport(layout.pos.col = 1))
circos.clear()
circos.par(start.degree = 0, gap.after = 10)

circos.heatmap(
  mat_t,
  col = col_fun,
  na.col = "grey80",
  cluster = TRUE,
  dend.side = "none",
  rownames.side = "outside",
  track.height = 0.5,
  rownames.font = 2,
  rownames.cex = 1.5,
  rownames.col = "black"
)

# Add sample names on outer track
circos.track(track.index = 2, panel.fun = function(x, y) {
  cn <- colnames(mat_t)
  n <- length(cn)
  if (CELL_META$sector.numeric.index == 1) {
    circos.text(rep(CELL_META$cell.xlim[2], n) + convert_x(1, "mm"),
                1:n - 0.5, cn,
                cex = 1.3, adj = c(0.5, 1), facing = "clockwise", font = 2)
  }
}, bg.border = NA)

upViewport()  # Exit heatmap

# Legend (right)
pushViewport(viewport(layout.pos.col = 2))
draw(lgd_combined, just = "left")
upViewport()

popViewport()  # Exit main layout
dev.off()

Can anyone please help in figuring out the issue?

[dariober]

Have you considered alternatives to a heatmap? My first try would be a line plot with labels (fat.1, fatt2., etc) on the x-axis, the value on y-axis and a line for A, B, C, D, E. It may become a very wide plot if you want the labels to be all readable, but maybe you don't need that and it could still be more space-efficient and readable than a heatmap. My issues with heatmaps is that the colour scale makes more difficult to see patterns and make comparisons relative to a regular line or dot plot.

[Yukta]

The need is a heatmap only, so I cannot opt for another figure. Thank you for responding though ^^

Answer 1

[Bastien Hervé]

Open an issue here

In circos.heatmap vignette, on the first heatmap, you can guess the row R71 in the block "a". With C10/C9 as positive, C3/C4/C5/C6 as negative and C1/C2/C7/C8 around 0.

On section 6.3 on the circlize heatmap, R71 is on the right side of the plot, third to last.

Guessing by the colors, one would expect C1/C2 next to the outside border and C9/C10 on the inside.

But in section 6.8 on the circlize heatmap, you can see that the labels are flipped as well.

[Yukta]

Yes, that's the issue, the column's information is getting flipped, but not the column headers. In my dataframe , the values of A are on the outside and the values of E are inside but the headers are vice-versa. I have opened an issue on the link you provided. Thank you for responding and giving your insights.