Skip to content

Loewe Additivity / Combination Index

The Combination Index (CI) is the most widely cited framework for quantifying drug synergy in pharmacology. It is based on Loewe additivity — the null hypothesis that two drugs act as dilutions of each other.

The Model

For a combination of drugs A and B at doses \(a\) and \(b\) producing observed effect \(E\), Loewe additivity requires:

\[ \frac{a}{A(E)} + \frac{b}{B(E)} = 1 \]

where \(A(E)\) and \(B(E)\) are the single-agent doses that produce the same effect \(E\). This sum is the Combination Index:

\[ \text{CI} = \frac{a}{A(E)} + \frac{b}{B(E)} \]
CI Interpretation
CI < 1 Synergy — combination is more potent than expected
CI = 1 Additivity — drugs act as dilutions of each other
CI > 1 Antagonism — combination is less potent than expected

Inverting the Hill Curve

\(A(E)\) and \(B(E)\) are found by analytically inverting the Hill equation:

\[ A(E) = \text{IC}_{50} \cdot \left(\frac{\text{top} - \text{bottom}}{E - \text{bottom}} - 1\right)^{1/n_H} \]

This is exact — no numerical root-finding needed.

When CI is Undefined (NaN)

The inversion requires \(\text{bottom} < E < \text{top}\) (strictly). For cells where the observed response is at or beyond the single-agent range, CI cannot be defined:

  • \(E \leq \text{bottom}\): complete inhibition — the drug combination exceeds any achievable single-agent effect
  • \(E \geq \text{top}\): no effect — neither drug produces measurable inhibition at these concentrations

These cells appear as grey (NaN) in the heatmap. Heavily grey heatmaps often indicate that concentration ranges for one drug are too extreme relative to its IC₅₀.

The implementation uses shared asymptotes effect_0 and effect_inf (for inhibition, these correspond to Top and Bottom in the notation above); inversion is valid only when \(E\) lies strictly between them, including activation assays where the two asymptotes are reversed relative to inhibition.

Slope mismatch

Loewe's sham-combination principle — "a drug combined with itself has CI = 1" — is rigorously defined only when both drugs share the same Hill slope. When the two fitted slopes disagree, the isobole equation above still produces a number, but part of that number reflects the shape mismatch rather than any real pharmacological interaction. This is a well-documented caveat of CI (Berenbaum, 1989; Foucquier & Guedj, 2015).

synfit flags this automatically on the synergy panel using the ratio of the two fitted Hill coefficients:

Severity Ratio max(h) / min(h) Meaning
(none) < 1.5, or both slopes ∈ [0.9, 1.1] Slopes matched closely enough, or both near the special Hill = 1 case where Loewe is well-behaved.
Yellow 1.5 ≤ ratio < 3.0 Theoretical foundation strained. CI is still informative but expect some bias.
Red ratio ≥ 3.0 Loewe geometry badly broken. Treat CI values as suspect until the slope mismatch is resolved.

What to do when you see this warning

Open the matrix analysis setup and pin one or both Hill slopes in the per-drug parameter rows (hill_a, hill_b):

  • To a known biological value if literature gives you a reliable Hill coefficient for this drug / assay.
  • To 1.0 as a neutral baseline — this is the "no cooperativity" assumption and makes Loewe and Bliss coincide most closely; use when you don't have a principled reason to prefer anything else.
  • Uncheck fit and set init to the chosen value; the joint fit will pin the slope and re-fit the remaining parameters.

Tying both slopes to a single shared parameter in the joint fit (rather than fixing them to the same number manually) is a separate planned feature.

Comparison with Bliss and HSA

Property Bliss HSA Loewe CI
Null hypothesis Probabilistic independence Better single agent Dose additivity
Requires fitted IC₅₀? Yes No Yes
Undefined cells? No No Yes (outside single-agent range)
Most cited in literature? Cell biology Drug discovery screens Pharmacology (Chou-Talalay)

Different models can disagree. When Bliss shows synergy but Loewe shows additivity, it often means the drugs have different maximal effects — Bliss is sensitive to this, Loewe is not.

Further Reading

Implementation details

Core function (synfit.loewe):

  • loewe_ci(conc_hor, conc_ver, c50_hor, c50_ver, hill_hor, hill_ver, mean_matrix, effect_0, effect_inf, asymmetry_hor=1.0, asymmetry_ver=1.0) → np.ndarray shape (n_ver, n_hor) — computes CI for each matrix cell. Calls _invert_hill() for both drugs, then CI[i,j] = conc_hor[j]/A(E) + conc_ver[i]/B(E). Returns NaN where the observed effect lies outside the open interval between the two asymptotes (handles inhibition and activation via min/max of effect_0 and effect_inf in _invert_hill).
  • _invert_hill(y, c50, hill, effect_0, effect_inf, asymmetry=1.0) → np.ndarray — analytical inversion of the 5-parameter Hill curve (collapses to the 4p formula when asymmetry = 1); returns NaN outside the valid response band.
  • slope_mismatch_warning(hill_hor, hill_ver) → dict | None — returns the structured warning payload shipped on the synergy panel. Thresholds: yellow at ratio ≥ 1.5, red at ratio ≥ 3.0; suppressed when both slopes lie inside [0.9, 1.1].

Visualization (synfit.plotting):

  • loewe_heatmap(mean_matrix, conc_horizontal, conc_vertical, c50_hor, c50_ver, hill_hor, hill_ver, effect_0, effect_inf) → PNG bytes — computes CI matrix, then calls matrix_heatmap() with colormap RdBu_r centred at CI=1 (vmin = 1 - lim, vmax = 1 + lim), color_reference=1.0 for a symmetric log colour scale around additivity, and colorbar label "Combination Index". NaN cells render as grey.