← Bioinformatics Series

Waterfall Plot

Ranking patients by best tumor change, then reusing the same visual grammar for molecular measurements.

In oncology, a waterfall plot usually means one bar per patient, sorted by best percent change in target-lesion burden. Negative values mean shrinkage. Positive values mean growth. The plot is popular because it keeps the individual patients visible instead of collapsing them into a single response rate.

The same ranked-bar grammar also works for molecular quantities such as expression or mutation burden. The important habit is to name the quantity being ranked, because a RECIST response waterfall and a molecular waterfall answer different questions.

CR
PR
SD
PD

Figure 1. RECIST-style waterfall for 60 synthetic patients. Bars show best percent change from baseline in the sum of target-lesion diameters. The -30% line marks the target-lesion threshold for partial response; the +20% line is a progression reference, but overall progression can also come from non-target lesions or new lesions. A bar crossing -30% is therefore a candidate response, not an ORR by itself.

Now reuse the ranking idea for the TP53CDKN1A relationship. Sorting by one variable and coloring by another exposes associations: a genotype that clusters at one end of the phenotypic spectrum is visible immediately.

TP53 mutant
TP53 wild-type

Figure 2. CDKN1A expression waterfall for the same 60 patients. The bars are sorted descending. TP53 mutant samples (red) cluster toward the low-expression end, as TP53 is a transcriptional activator of CDKN1A.

CDKN1A (p21) is a cell cycle inhibitor. Mutated TP53 cannot activate CDKN1A, decreasing expression and allowing faster cell division.

Sorting by dimension

A cohort can be reordered by different measures. Tumor mutation burden (TMB) is often elevated in mutant tumors due to impaired DNA repair — a second axis that tells a different story on the same patients.

Figure 3. Reshuffling the cohort. TP53 mutants cluster low on CDKN1A but high on TMB.

Threshold explorer

Dragging a threshold classifies patients (e.g., CDKN1A-low) and computes mutation enrichment in that group, bridging visual pattern recognition and formal testing.

Threshold (log₂)
Enrichment (Odds Ratio)
Fisher's Exact p-value

Figure 4. Interactive thresholding. Drag the dashed line to classify patients; statistics update to show the association with TP53 status. This is exploratory, so the p-value should be treated as descriptive unless the cutoff was pre-specified or validated elsewhere.