What is Design of Experiments (DOE)?

Design of Experiments (DOE) is a systematic, statistical approach to planning experiments that vary multiple factors simultaneously. Unlike one-factor-at-a-time (OFAT) testing, DOE reveals interaction effects between factors, requires fewer total runs, and enables mathematical modeling of the factor-response relationship. In bioprocess development, DOE is essential for optimizing parameters like pH, temperature, feed rate, and dissolved oxygen.

When should I use Full Factorial vs Fractional Factorial design?

Full Factorial tests every combination of factor levels — ideal for 2-4 factors where you need complete interaction information (2^k runs). Fractional Factorial (Resolution III-V) uses a strategically chosen subset of runs — use it for 5+ factors in screening studies where you can sacrifice higher-order interaction estimates. For example, a 2^(5-1) fractional factorial tests 5 factors in only 16 runs instead of 32.

What is the difference between CCD and Box-Behnken designs?

Both are response surface designs for fitting quadratic models. Central Composite Design (CCD) uses factorial points plus axial (star) points at a distance alpha from the center — it can estimate all quadratic terms and runs at 5 levels per factor. Box-Behnken uses midpoints of edges of the factor space — it requires 3 levels, uses fewer runs for 3-4 factors, and avoids extreme corners. Choose CCD when you can test at extreme settings; choose Box-Behnken when extreme combinations are infeasible or risky.

How many center points should I include in my DOE?

Include 3-5 center point replicates for most designs. Center points serve three purposes: they provide a direct estimate of pure experimental error (repeatability), enable a lack-of-fit test to check if a linear model is adequate (curvature detection), and do not affect the estimation of main effects or interactions. For screening designs (Plackett-Burman), 3 center points are typical. For response surface designs (CCD, Box-Behnken), 3-6 center points are recommended.

What is a Plackett-Burman design used for?

Plackett-Burman designs are Resolution III screening designs used to identify the most important factors from a large set of candidates (up to 7-47 factors). They use the minimum number of runs (multiples of 4) to estimate main effects only — interaction effects are confounded. Use Plackett-Burman as a first screening step in bioprocess development to narrow down from 8-12 potential factors to the 3-5 most significant ones, which you then optimize with a response surface design.

How do I choose the right DOE design for bioprocess optimization?

Follow this decision tree: (1) Screening many factors (5+)? Use Plackett-Burman or Fractional Factorial. (2) Characterizing 2-4 factors with interactions? Use Full Factorial. (3) Optimizing 2-5 factors to find the best settings? Use CCD or Box-Behnken. (4) Screening and optimizing in one step with 3-6 factors? Use Definitive Screening Design. For bioprocess applications, a typical workflow is Plackett-Burman screening → CCD optimization → confirmation runs at the predicted optimum.

DOE Experiment Generator

Design of Experiments — Run Table Generator

How to use: set your factors and design type, generate the experiment matrix, then review and export the run table. Follow the numbered steps above.

1Define Objective (what to measure & optimise)

Response	Unit	Goal	Target	Cap

2Bioprocess Preset

2Factors (2-8 supported)

Name	Low (-1)	High (+1)	Unit	HTC

3Which design should I use?

Answer two questions. We recommend a design and show how many runs each option needs. No statistics background needed.

Your goal

Extreme settings risky?

3Design Type

Center Points

Randomize

D-Optimal runs

Number of runs to generate. Leave blank for an automatic minimum. The algorithm chooses the most informative settings for a quadratic model.

Power & sample size Pro

How many runs you need to reliably detect the effect you care about.

Smallest effect to detect

Run-to-run SD (σ)

Significance (α)

Power

4Run trials

3Design Summary

3Visual Design Matrix

High (+1) Low (−1) Center (0) Axial (+α) Axial (−α)

3Factor Levels Across Runs

4Run Table

Quick fill:

5Analyze Interactions

Response:

How to read this panel: Each bar in the Pareto chart shows how strongly that factor (or interaction) affects your response. Bars to the right of the dashed red line are statistically significant (p<0.05) — those are the levers worth tuning. Bar colours indicate term type: teal = main effect, blue = 2-factor interaction, purple = quadratic curvature. The Effects Table below gives precise numbers and the R² tells you how well the model fits (closer to 100% = better).

6Identify Optimum

How to read this panel:

Best Measured Run — the actual run from your data with the best score. Use these settings to reproduce a result you've already seen.
Predicted Optimum — where the fitted model says the true peak is. This may not be one of your runs — the model interpolates between data points. To validate, run 2–3 confirmation experiments at the predicted settings.
Multi-Response Best Compromise — when you have multiple responses with different goals (e.g. maximise titre AND minimise HCP), this finds the single set of factor values that scores well on all of them simultaneously (Derringer-Suich desirability).

Prediction profiler Pro — move each factor to see the response respond

Response:

Each curve auto-scales to its own range; compare the shape, not the height.

DOE Pro+ Pro+ Bayesian optimisation & design augmentation

Enter the results you have so far in the Step 5 table, then let the tool suggest the next best experiment(s), or add runs to grow an existing design. Methods are implemented from published algorithms (see the sources note below).

Suggest next experiment (Bayesian optimisation)

Fits a Gaussian-process model to your measured runs and proposes the run(s) most likely to improve the chosen response (Expected Improvement).

Optimise response

How many suggestions

Strategy

Exploration

Augment this design (add runs, keep what you've done)

Grow your current design without discarding any runs you've already completed. Note: changing a factor, the design type, or loading a preset rebuilds the base design and clears added or suggested runs, so augment last.

Grow to response surface

Add N optimal runs (D-optimal)

Methods: Gaussian-process regression (Rasmussen & Williams, Alg. 2.1) with Expected Improvement (Jones et al. 1998) and a Matérn-5/2 kernel (Snoek et al. 2012); D-optimal augmentation by coordinate exchange (Meyer & Nachtsheim 1995); response-surface growth (Montgomery, Ch. 11). Full sources.

Design of Experiments in Bioprocess Development

Design of Experiments is a cornerstone of Quality by Design (QbD) approaches mandated by ICH Q8 for pharmaceutical development. In bioprocess engineering, DOE is used at every stage: screening critical process parameters (CPPs) during early development, optimising conditions during process characterisation, and defining the design space for regulatory filings. A well-designed experiment can save weeks of bench time compared to one-factor-at-a-time approaches while providing far richer information about factor interactions and process robustness.

Common bioprocess DOE applications include CHO fed-batch optimisation (pH, temperature, dissolved oxygen, feed rate, osmolality), E. coli expression screening (inducer concentration, induction temperature, induction OD, post-induction time), downstream purification optimisation (loading density, flow rate, buffer pH, salt concentration), and media formulation studies. This tool now handles the full pipeline from design to predicted optimum in your browser — you can still export the run table to external statistical software if your workflow prefers it.

How to Use the DOE Experiment Generator

This free Design of Experiments (DOE) generator is a complete platform — design, analyse, and identify the optimum in one tool, without paying for JMP, Minitab, or Design-Expert. The workflow follows the standard 6-step DOE process visible in the stepper at the top of the page:

Define objective — name the responses you'll measure (titre, viability, productivity, HCP, etc.) and choose a goal for each: maximise, minimise, or hit a target value. Up to four responses can be optimised simultaneously.

Choose factors & ranges — pick a bioprocess preset (CHO Fed-Batch, E. coli Expression, Pichia, Chromatography, Cell Culture Media, UF/DF, AAV Production) or define your own 2–8 factors with low/high levels and units.

Design experiment — pick one of six design types. Each tab shows the live run count for your current factor set so you can balance information against lab effort: Full Factorial (2^k), Fractional Factorial (2^k-p), Plackett-Burman, Central Composite (CCD), Box-Behnken, or Definitive Screening (DSD).

Run trials — export the randomised run sheet to CSV or print it. Record measured responses directly in the Run Table (cells are editable), paste a column from Excel, or click Demo data to see how the analysis works before you have real data.

Analyze interactions — fit an OLS regression model, see a plain-language summary, the Effects Pareto with significance threshold, an Effects Table with p-values and star ratings, residual diagnostics (residual-vs-fitted + Q-Q plot), and full ANOVA with lack-of-fit decomposition when replicates exist.

Identify optimum — see which actual run scored best, the analytically-solved per-response optima, the multi-response best compromise via Derringer-Suich desirability, and a contour plot of any two factors with the optimum marked. Add 3 confirmation runs at the predicted settings with a single click to validate the prediction in the lab.

Built-in design types: Full Factorial for complete characterisation of 2-4 factors, Fractional Factorial for efficient screening of 4-8 factors, Plackett-Burman for main-effect screening, Central Composite Design (CCD) for response surface modelling, Box-Behnken for 3-level designs that avoid extreme corners, and Definitive Screening Design (DSD) for combined screening + RSM in minimal runs.

All analysis runs client-side in your browser — no data leaves your machine. CSV export is still available at any stage so you can also feed the run table to R, Python (statsmodels), JMP, Minitab, or Design-Expert if you prefer.

DOE Experiment Generator

Design of Experiments in Bioprocess Development

How to Use the DOE Experiment Generator

DOE Articles

Start here: DOE fundamentals

Choosing a design type

Applications, regulatory & software

Frequently Asked Questions

Resources

DOE Experiment Generator

Design of Experiments in Bioprocess Development

How to Use the DOE Experiment Generator

DOE Articles

Start here: DOE fundamentals

Where Do I Even Start?

DOE vs OFAT

DOE for Bioprocess Optimization

Which DOE Design Should I Use?

How Many Experiments Does a DOE Need?

How to Read Your DOE Results

Choosing a design type

Full Factorial Design

Fractional Factorial & Aliasing

Plackett-Burman Designs

Definitive Screening Designs (DSD)

Response Surface Methodology

Mixture Design DOE

Applications, regulatory & software

DOE for Cell Culture & Fermentation

Media Optimization with DOE

ICH Q8 DOE & Design Space

ICH Q2 DOE for Analytical Methods

CPP/CQA Mapping & Design Space

Free Alternatives to JMP & Minitab

Frequently Asked Questions

Resources