Skip to content

Drug combination synergy analysis online

A drug combination screen measures the response to two drugs across a grid of concentrations, a dose-response matrix or checkerboard. The question is whether the combination does more than you would expect from each drug alone (synergy), less (antagonism), or about the same (additivity).

"Expected" needs a model, and the four standard models give different answers. synfit computes all four from one upload, Bliss independence, Loewe additivity (combination index), HSA and ZIP, so you can see whether they agree before you claim synergy. For which one to trust, see Bliss vs Loewe vs HSA vs ZIP.

Try it without an account

Open the demo: the sample datasets include combination matrices with no interaction, synergy and strong synergy, plus a drug combined with itself.

1. Lay out the matrix

Each replicate is one CSV file holding the response grid only, no headers and no concentrations:

1.02,0.98,0.85,0.45,0.12
0.97,0.94,0.80,0.41,0.10
0.88,0.85,0.72,0.35,0.08
0.52,0.48,0.38,0.18,0.04
0.15,0.12,0.09,0.05,0.02

Columns are the concentrations of the horizontal drug, rows those of the vertical drug. You enter both concentration series on the upload screen, by hand or with the serial-dilution generator. Details: CSV formats.

Two design rules matter more than anything else:

  • Include a zero row and a zero column. Wells with only one drug are the single-agent dose-response curves. Every synergy model compares the combination wells against these, so without them there is nothing to compare against.
  • Cover each drug's own IC₅₀. If a drug has no visible effect within its tested range, its curve cannot be fitted, and every synergy score built on it is unreliable.

A 6×6 or 8×8 grid with 2–4 replicates is typical. Small grids give little to fit: a single-agent curve needs several non-zero doses spread around its IC₅₀, and ZIP needs at least three non-zero doses in each row and column, so a 2×2 or 3×3 design returns no ZIP score.

2. Upload and fit

  1. Upload → Matrix. Add one file per replicate, name both drugs, set whether each is an inhibitor or activator, choose the unit, and enter the concentrations for each column and row. Both drugs must act in the same direction for synergy scores to be defined.
  2. Check the heatmap preview. All replicates share one colour scale, so a plate that looks different from the others stands out.
  3. New analysis. Choose the error model (Auto is fine) and, per drug, the 4- or 5-parameter curve model.
  4. Fit. synfit fits both single-drug curves jointly (each drug its own potency and slope, shared top and bottom plateaus), then computes the four reference surfaces and scores.

3. Read the results

Output What it tells you
Per-drug parameters IC₅₀ (or EC₅₀) and Hill slope for each drug, from the single-agent edges
Bliss heatmap Deviation from independent action. Details
HSA heatmap Deviation from the better single drug at that dose. Details
Loewe CI heatmap Combination index: < 1 synergy, 1 additive, > 1 antagonism. Details
ZIP heatmap δ-score from per-row and per-column re-fits, as in SynergyFinder. Details
Deviation / Reference toggle Shows the expected (no-interaction) surface itself instead of the deviation. Details

Colours: in all four heatmaps blue means synergy, red means antagonism and white means the combination behaves as the model expects. For Bliss, HSA and ZIP the numbers are deviations (negative = synergy); for Loewe it is the CI on a log colour scale centred at 1.

Look for regions, not single cells. A block of neighbouring cells pointing the same way is a finding; an isolated coloured cell is usually noise. Compare replicates in the preview if one cell drives the picture.

4. Warnings worth reading

synfit flags situations where a score is likely to mislead:

  • Bliss and Loewe disagree. A banner appears when their overall direction or size differs. This is common and informative, not an error: see why the metrics can disagree.
  • Hill slope mismatch. Loewe CI assumes comparable slopes; when the two drugs' Hill coefficients differ by 1.5× or more, the CI panel warns you (slope mismatch).
  • Undefined cells. Loewe CI cannot be computed where the response lies outside the range the single-drug curves span (complete effect, or no effect at all); those cells are grey. ZIP lists rows or columns whose re-fit failed. Neither is shown as "additive".
  • Opposite directions. If one drug inhibits and the other activates, there is no shared expectation and synfit does not show synergy heatmaps.

5. Report it

  • Download any heatmap or curve as PNG or SVG.
  • Send a read-only share link to a collaborator (sharing).
  • State which model each claim rests on, and whether the others agree. "Synergistic by Bliss and ZIP, additive by Loewe" is a more useful sentence than "synergistic".

Frequently asked

Which synergy score should I report? The one whose assumption fits your biology, and say whether the others agree. The comparison guide walks through the choice.

Is this the same ZIP as SynergyFinder? It follows the published definition (Yadav et al., 2015) that SynergyFinder uses by default, with two differences in presentation and one in fitting. synfit reports δ as a fraction with negative = synergy; SynergyFinder reports percentage points with positive = synergy, so synfit's −0.15 corresponds to roughly +15 in SynergyFinder. synfit also fits the single-agent curves jointly with an explicit noise model (and optionally a 5-parameter curve), so the values can differ, sometimes noticeably on noisy or poorly sampled matrices. Compare the pattern across the matrix rather than individual numbers.

Do I need to normalise to % inhibition first? No. The fitted top and bottom plateaus define 0 % and 100 % effect, so raw signal (luminescence, absorbance, counts) works.