Using DOE to Optimize ELISA and Immunoassay Development

Part of our complete guide to DOE for biologists.

Assay development usually goes one reagent at a time: try three coating concentrations, pick the best, then try three blocking buffers against that fixed coating concentration, pick the best, then move to detection conjugate dilution. Each step looks efficient. What it misses is that the best blocking buffer at one coating concentration isn't necessarily the best one at another — the variables interact, and a one-at-a-time search can settle on a combination that's good, not optimal, without ever revealing it missed something better.

The variables that usually matter

• Continuous: Antibody coating concentration, detection conjugate dilution, incubation time and temperature.
• Categorical: Blocking agent (BSA vs. casein vs. a commercial blocker), substrate choice.

Mixing continuous and categorical variables in one design is normal — you don't need to run separate experiments for each type.

What you're actually optimizing

Most immunoassay development is optimizing signal-to-noise ratio: strong specific signal, low background. Sometimes dynamic range matters too — the assay needs to distinguish concentrations across the range you'll actually see in samples, not just produce the single strongest signal at one concentration. If both matter, set both as responses and optimize them together rather than picking one and hoping the other comes along for free.

Why this usually needs curvature, not just a straight line

Dilution series are a textbook case of a true optimum sitting inside your tested range: too concentrated and background climbs, too dilute and specific signal drops below detection. A design that only tests a low and a high level can tell you the general direction, but not where the peak actually is. If your screening result comes back with curvature detected, that's the design telling you it's found evidence of exactly this kind of peak — and BioDOE can extend the design into a full response surface to pin it down, rather than making you restart with more levels chosen by guesswork.

Sizing the experiment to your plate variability

Plate-to-plate and lot-to-lot variability is real background noise in any immunoassay. Before running a full optimization, it's worth checking: given how much your signal-to-noise ratio normally varies between identical plates, and the smallest improvement worth chasing, does your planned number of runs actually have a shot at detecting it?

Reading the result

The significance table tells you which reagent concentrations and conditions genuinely move signal-to-noise, versus which ones you can stop optimizing and just fix at a convenient value. If curvature is present and you optimize, the recommended settings come with a predicted signal-to-noise value — worth confirming with a few verification runs before it becomes your new standard protocol.

Start free and run your next assay optimization as a single coordinated design instead of a chain of one-at-a-time comparisons.