Abstract
We propose a fold change visualization that demonstrates a combination of properties from log and linear plots of fold change. A useful fold change visualization can exhibit: (1) readability, where fold change values are recoverable from datapoint position; (2) proportionality, where fold change values of the same direction are proportionally distant from the point of no change; (3) symmetry, where positive and negative fold changes are equidistant to the point of no change; and (4) high dynamic range, where datapoint values are discernable across orders of magnitude. We outline a new transform and visualization, named mirrored axis distortion of fold change (MAD-FC), that extends a linear visualization of fold change data to exhibit readability, proportionality, and symmetry. We illustrate the use of MAD-FC with biomedical data and argue that MAD-FC plots may be a more useful visualization than log or linear plots for applications that require a limited dynamic range (approximately ±2 orders of magnitude).
Introduction
In bioinformatics and genomics, researchers frequently express results as "fold change"—the ratio of an experimental group to a control. Despite its ubiquity, visualizing fold change presents a persistent dilemma. Standard linear plots are intuitive for positive changes but compress all down-regulation (negative changes) into the narrow interval between 0 and 1. Conversely, logarithmic plots (such as Log2) achieve symmetry—ensuring a 2-fold increase and a 2-fold decrease are equidistant from the baseline—but they sacrifice proportionality. In a log plot, the visual distance between a 2-fold and 4-fold change is identical to the distance between a 4-fold and 8-fold change, even though the latter represents a significantly larger raw biological difference.
MAD-FC (Mirrored Axis Distortion of Fold Change) is introduced as a "missing middle" visualization technique. It combines the symmetry of log plots with the proportional intuition of linear scales, ensuring that the visual distance on a graph accurately reflects the magnitude of biological change in both directions.
The MAD-FC Framework
The MAD-FC transform is a mathematical process applied to data before plotting, consisting of three primary steps:
- Mirroring ($f_M$): To achieve symmetry, negative fold changes (values < 1) are transformed into their negative reciprocals. For example, a 0.25-fold change (a 4-fold decrease) becomes -4. This creates a symmetric scale but leaves a "dead zone" between -1 and 1 where no data can exist.
- Contraction ($f_C$): To remove this dead zone and create a continuous scale, the entire axis is shifted toward zero by one unit. This ensures that a fold change of 1 (no change) sits at the origin (0).
- Axis Relabeling: To maintain readability, the authors decouple the mathematical transform from the visual labels. The axis is labeled with familiar fold-change values (e.g., 1/4, 1/2, 1, 2, 4) while the points themselves are positioned according to the undistorted MAD-FC geometry.
Comparative Advantages
Preservation of Distribution and Uncertainty
One of the most significant insights provided by MAD-FC is its treatment of data distributions. In log plots, groups with the same standard deviation in fold-change units appear to have different widths depending on their distance from the origin; groups further away appear "compressed." MAD-FC is the only method that preserves the visual shape of the distribution regardless of the mean. This property extends to uncertainty; in simulated tests with identical 95% confidence intervals, MAD-FC was the only visualization where error bars remained a consistent length across all magnitudes.
Honest Summaries in High-Throughput Data
Log plots can provide a misleadingly crowded impression of high-magnitude changes, often making "top" candidates in a volcano plot appear closer to the maximum observed change than they truly are. MAD-FC provides a more "honest" visual summary, emphasizing the largest changes more effectively. This is particularly useful in heatmaps, where log mappings tend to saturate quickly, whereas MAD-FC color gradients better highlight the most extreme outliers.
Applications and Datasets
The utility of MAD-FC was validated using a variety of real-world biomedical datasets, including:
- RNA-Seq: Analyzing differentially expressed genes in airway smooth muscle cells.
- Proteomics: Mapping protein phosphorylation in HER2-positive breast cancer cells.
- The Cancer Genome Atlas (TCGA): Visualizing mRNA expression across breast, ovarian, and lung cancers.
- Interactomics: Comparing Ubiquitin-protein interactors via heatmaps.
Limitations and Scope
MAD-FC is not intended to replace log plots in all scenarios. Because it maintains a linear relationship between distance and magnitude, it cannot accommodate the massive dynamic ranges (spanning 4 or more orders of magnitude) that log scales handle with ease. The authors suggest MAD-FC is most effective for data within approximately ±2 orders of magnitude (or ±8 units in log2 space), which encompasses the vast majority of differential expression and proteomics studies.
Conclusion
MAD-FC offers a robust alternative for researchers who require a visualization that is symmetric, proportional, and highly readable. By preserving the visual integrity of distributions and confidence intervals, it allows for a more intuitive interpretation of biological magnitude than traditional log or linear scales.
Summary updated