為什麼長條圖+errorbar 該退役？
Weissgerber et al. 2015 在 PLOS Biology 的「Beyond Bar and Line Graphs」一文成為里程碑——同一組平均值 + SEM 可能對應截然不同的資料分布（單峰、雙峰、含 outlier）。務必顯示每個資料點。Lord et al. 2020 (JCB) 提出 SuperPlots：以顏色標示不同生物重複的細胞層級資料，同時呈現技術與生物變異。
Tufte 的「data-ink ratio」與 Wilke 2019 Fundamentals of Data Visualization 強調：去除多餘裝飾，用最少 ink 表達最多資訊。color-blind 友善的調色板（viridis 與 Okabe-Ito 8 色）已是現代期刊要求。
Weissgerber et al. 2015 (PLOS Biology) "Beyond Bar and Line Graphs" became a manifesto — identical mean + SEM can hide unimodal, bimodal, or outlier-laden distributions. Always show every data point. Lord et al. 2020 (JCB) introduced SuperPlots: color-code biological replicates and show cell-level + replicate-level variation together.
Tufte's "data-ink ratio" and Wilke 2019 Fundamentals of Data Visualization stress: remove chart junk, maximize information per ink. Color-blind-safe palettes (viridis, Okabe-Ito 8 colors) are now journal requirements.
一、依資料類型選圖
| 資料類型 | 推薦圖表 | 不推薦 |
|---|---|---|
| 單組連續分佈 | 直方圖+density | bar + SEM |
| 多組比較（小樣本） | 點圖 + 95% CI | bar + SEM |
| 多組（大樣本） | violin + boxplot, ridge | pie chart |
| 兩變量關係 | 散點+平滑+r | 3D 散點 |
| 多批次組學 | heatmap (cluster), PCA | 瀑布圖 |
| 差異基因總覽 | volcano | bar of top genes |
| ≥4 集合交集 | UpSet plot | Venn diagram |
| 統合分析 | forest plot | 分散 bar |
同一組資料 → 三種觀感
切換按鈕，觀察 bar+SEM、box plot、dot plot 對同一組資料的呈現差異。重點：bar+SEM 隱藏了 outlier 與雙峰，dot plot 全部攤開。
Toggle the buttons — see how bar+SEM, boxplot, and dot plot present the same data. Bar+SEM hides outliers and bimodality; dot plot shows it all.
兩組 n=20
二、SuperPlot 與 Volcano 範例
library(ggplot2); library(dplyr); library(ggrepel); library(viridis) # 1. SuperPlot — 顏色標示生物重複 sp <- df %>% group_by(replicate, group) %>% summarise(rep_mean = mean(value), .groups="drop") ggplot(df, aes(x=group, y=value, color=factor(replicate))) + geom_jitter(width=.15, alpha=.4, size=1.5) + geom_point(data=sp, aes(y=rep_mean), size=5, shape=95) + stat_summary(fun=mean, geom="crossbar", width=.4, color="black") + scale_color_viridis(discrete=TRUE, option="D") + theme_classic() + labs(color="Bio rep") # 2. Volcano plot with labels res_df$sig <- case_when(res_df$padj < .05 & abs(res_df$lfc) > 1 ~ "hit", TRUE ~ "ns") ggplot(res_df, aes(lfc, -log10(padj), color=sig)) + geom_point(alpha=.5) + geom_text_repel(data=filter(res_df, sig=="hit"), aes(label=gene), max.overlaps=20) + geom_vline(xintercept=c(-1,1), lty=2) + geom_hline(yintercept=-log10(.05), lty=2)
import seaborn as sns, matplotlib.pyplot as plt # 1. SuperPlot 風格 fig, ax = plt.subplots(figsize=(5,4)) sns.stripplot(data=df, x="group", y="value", hue="replicate", palette="viridis", jitter=.15, alpha=.4, ax=ax, size=4) rep_mean = df.groupby(["replicate","group"])["value"].mean().reset_index() sns.stripplot(data=rep_mean, x="group", y="value", hue="replicate", palette="viridis", marker="_", size=30, jitter=False, ax=ax, legend=False) sns.pointplot(data=df, x="group", y="value", color="k", errorbar=("ci",95), ax=ax, markers="D") # 2. Volcano fig, ax = plt.subplots() ax.scatter(res.lfc, -np.log10(res.padj), c="#ccc", s=8, alpha=.5) hits = (res.padj < .05) & (res.lfc.abs() > 1) ax.scatter(res.lfc[hits], -np.log10(res.padj[hits]), c="#3730a3", s=14) ax.axhline(-np.log10(.05), ls="--", c="#888"); ax.axvline(1, ls="--", c="#888"); ax.axvline(-1, ls="--", c="#888")
三、視覺化的雷區
❌ 3D 圓餅 / 3D 長條
視角扭曲比例。Tufte 與所有現代視覺化教科書一致反對。
3D distorts proportions. Universally rejected by modern viz textbooks.
❌ 軸不從 0 開始
誇大微小差異。長條圖必須從 0 開始；點圖可截斷但必須明顯標示。
Exaggerates trivial differences. Bar charts must start at 0; dot plots may truncate but with clear indication.
❌ 紅綠 heatmap
對最常見的紅綠色盲不友善。改用 viridis、RdBu、Okabe-Ito。
Unfriendly to red-green colorblind viewers. Use viridis, RdBu, or Okabe-Ito.
❌ * ** *** 不註明 p
現代期刊要求附 exact p。星號可保留為視覺指引，但 p 值必須在圖例或表格內。
Modern journals require exact p-values. Asterisks may stay as visual cues, but report exact p in legend or table.
🎯 章末小測驗
1. Weissgerber 2015 主要建議？
2. ≥4 集合交集用？
3. 色盲友善調色板？