Overview

Biological data visualization bridges the gap between raw quantitative data and human interpretation. Modern high-throughput technologies generate datasets with thousands to millions of measurements, making effective visualization essential for hypothesis generation, quality control, and communication. Good visualizations reveal patterns, outliers, and relationships that summary statistics alone cannot capture, while poor ones can mislead and obscure. The field draws on principles from perceptual psychology, graphic design, and computer science to create representations that align with how the human visual system processes information.

Key Concepts

Data-ink ratio and chartjunk are foundational design considerations. The data-ink ratio, popularized by Edward Tufte, measures the proportion of a graphic’s ink devoted to displaying data versus decorative elements. Maximizing this ratio produces clearer figures. Color encoding must account for color vision deficiencies; tools like ColorBrewer help select perceptually uniform and accessible palettes. Scales and transformations: log scales, normalized axes, and multidimensional scaling, can reveal structure hidden in raw measurements. Overplotting in dense scatter plots is addressed through transparency, hexagonal binning, or kernel density estimation.

Applications

Visualization permeates every stage of biological research. During exploratory analysis, scatter plots and heatmaps of DNA microarray and gene expression data identify differentially expressed genes. In flow cytometry, bivariate dot plots and density plots reveal cell populations based on surface markers. Structural biologists use ribbon diagrams and surface representations to communicate protein structure. High-dimensional data from single-cell sequencing and proteomics increasingly relies on dimensionality reduction plots such as t-SNE and UMAP for visualization.

Plot Types and Best Practices

For genomics data, the most common plot types each serve a specific purpose. Volcano plots display differential expression results: the x-axis shows log₂ fold change, the y-axis shows -log₁₀ adjusted p-value, with horizontal lines marking significance thresholds and vertical lines marking fold-change cutoffs. Points above both thresholds are significantly differentially expressed genes. MA plots (mean vs. ratio) are preferred for RNA-seq data because they reveal intensity-dependent bias: the x-axis shows average expression, the y-axis shows log fold change, and genes with low average expression exhibit higher variance. Heatmaps with hierarchical clustering are ideal for visualizing expression patterns across samples: rows are genes, columns are samples, and the color intensity represents expression level (typically Z-scores). Annotate columns with phenotypic metadata (treatment, time point, tissue) and rows with functional categories to reveal group-specific patterns. For transcriptomics, bar plots of normalized read counts with individual data points overlaid show gene-level expression changes while communicating biological variability. For genomics, manhattan plots display GWAS results across chromosomes, highlighting loci that exceed the genome-wide significance line. In all cases, follow design principles: use the smallest data-ink ratio that communicates the result, avoid 3D effects that distort perception, choose sequential color schemes for continuous variables and qualitative schemes for categorical variables, and always label axes clearly with units. A well-designed MA plot, for instance, should show red points for significant genes, display the running mean as a blue loess curve, and include the total number of up- and down-regulated genes in an inset label, enabling the reader to assess both the biological result and the technical quality of the experiment in a single glance.