Spearman Rho
Also known as: Spearman ρ, Spearman rank correlation, rho
A rank correlation coefficient (−1 to +1) that measures whether two variables agree in order, not magnitude. The primary metric for variant effect benchmarks.
Source: Spearman C. 'The proof and measurement of association between two things.' Am J Psychol 1904.
Spearman rank correlation (ρ) measures how well the ordering of one variable predicts the ordering of another. Unlike Pearson correlation, it makes no assumptions about linearity — only that the ranks agree.
Interpretation
| ρ value | Meaning |
|---|---|
| +1.0 | Perfect agreement in ranking |
| +0.7 | Strong agreement |
| +0.5 | Moderate agreement |
| 0.0 | No relationship |
| −1.0 | Perfect inverse ranking |
Why It’s the Right Metric for Variant Effects
When benchmarking computational variant predictors against deep mutational scanning (DMS) data, Spearman rho is preferred over Pearson correlation because:
- DMS fitness scores and ESM-2 log-likelihood scores have different scales and distributions
- What matters is whether the model ranks deleterious variants above tolerated ones — not whether the raw numbers are proportional
- Threshold choices (e.g., “damaging” vs “benign” cutoffs) don’t affect the rank-based metric
A Spearman rho of 0.679 means ESM-2’s ordering of CYP2C9 variants closely matches the experimentally measured ordering — even if the absolute score scales are completely different.
Comparison to Categorical Agreement
“Categorical agreement” (e.g., 6.8% for CYP2C9) measures how often two binary labels agree at a specific threshold. This metric is highly sensitive to where you draw the cutoff and can appear low even when the rank correlation is strong. For variant effect benchmarks, Spearman rho is the more informative metric.