What is a definitive screening design?
A definitive screening design is a three-level experimental design, introduced by Bradley Jones and Christopher Nachtsheim in 2011, that estimates the main effects of continuous factors completely free of two-factor interactions while also estimating quadratic (curvature) effects. It does all of this in roughly 2k+1 runs for k factors — for example, 13 runs for 6 factors. That combination of properties is unusual, and it is why the DSD has become a favourite for efficient process development.
The dsd earns the word "definitive" from a specific guarantee: unlike a resolution III screen, no main effect is aliased with any two-factor interaction, and no quadratic effect is aliased with any main effect. Two-factor interactions are only partially aliased with each other, not with mains. In practice this means a large main effect you find is genuinely a main effect, not an interaction in disguise — the ambiguity that always haunts Plackett-Burman results.
Each factor takes three coded levels: low (−1), center (0), and high (+1). The center level is the source of the design's curvature-detecting power, because a factor must be tested at three points before any quadratic term can be estimated. This places the DSD firmly between pure screening and full response-surface optimization in the design-of-experiments toolkit.
DSD vs screening-then-RSM
The conventional route to an optimized process runs two separate experiments: a two-level screen to find the critical factors, then a response-surface design (central composite or Box-Behnken) on the survivors to map curvature. A definitive screening design compresses both steps into one. It screens and begins optimizing at the same time, because it carries both the resolution to rank factors and the three levels to model curvature.
The payoff is run economy. Consider six factors. The classic path is a 12-run Plackett-Burman screen followed by a ~20-run central composite design on the three factors that survive — about 32 runs across two studies, plus the calendar time to analyze the first before designing the second. A DSD does it in 13 runs, in one study.
There is a caveat. Because the DSD spends so few runs, its power to resolve many active two-factor interactions is limited — if your process turns out to be dominated by interactions among many factors, you may still need an augmentation run set. But for the common case where a handful of main effects and one or two curvatures dominate, the DSD lands the answer in one pass.
When DSD wins (4–8 factors, few runs)
A definitive screening design is the right choice when you have 4 to 8 continuous factors, you suspect the response has an optimum (curvature) somewhere in the design region, and every run is expensive. That profile describes most bioprocess development perfectly — bioreactor runs cost days and reagents, so a method that screens and optimizes together is worth a great deal.
The design has clear boundaries. It needs continuous factors, so it is a poor fit when most of your factors are categorical (media type, clone, vendor). It needs at least four factors to be efficient; with two or three, a full factorial design with center points is simpler and just as good. And it assumes a sparse model — a few dominant effects rather than a dense web of interactions.
- Use a DSD when: 4–8 continuous factors, curvature suspected, runs costly, you want screen + optimize in one study.
- Use Plackett-Burman when: many factors (8–20), pure screening only, curvature not yet a concern.
- Use a full factorial when: only 2–4 factors and you need every interaction.
- Use a central composite / Box-Behnken when: the critical factors are already known and you need a precise optimum.
How a DSD is built (the run sheet)
A definitive screening design is constructed from a conference matrix folded over on itself, plus a single center run. For k factors this gives 2k+1 runs. Each of the fold-over rows has exactly one factor at its center level and the rest at their extremes, and every row is paired with its mirror image — the structure that delivers the clean separation of main effects from interactions.
| Run | A | B | C | D | E | F |
|---|---|---|---|---|---|---|
| 1 | 0 | + | + | + | + | + |
| 2 | + | 0 | + | − | − | + |
| 3 | + | + | 0 | + | − | − |
| 4 | + | − | + | 0 | + | − |
| 5 | + | − | − | + | 0 | + |
| 6 | + | + | − | − | + | 0 |
| 7 | 0 | − | − | − | − | − |
| 8 | − | 0 | − | + | + | − |
| 9 | − | − | 0 | − | + | + |
| 10 | − | + | − | 0 | − | + |
| 11 | − | + | + | − | 0 | − |
| 12 | − | − | + | + | − | 0 |
| 13 | 0 | 0 | 0 | 0 | 0 | 0 |
Notice the pattern: runs 1–6 are the conference matrix, runs 7–12 are their exact mirror images (every sign flipped), and run 13 is the all-center run. Each factor sits at its center level in exactly two runs and at an extreme in the rest, so every factor is measured at all three levels. Analysis is by regression — typically forward selection respecting effect hierarchy — rather than the simple column arithmetic used for two-level designs.
Definitive screening design DOE bioprocess optimization
For biological process optimization, the DSD is often the single most efficient design available, and "definitive screening design doe bioprocess optimization" captures exactly why teams adopt it: it answers which factors matter and roughly where the optimum lies, in one short campaign. Recent work has paired DSDs with high-throughput micro-bioreactors to develop cell-culture processes in a fraction of the usual runs.
Consider a six-factor CHO process: A = temperature, B = pH, C = dissolved oxygen, D = feed rate, E = glucose setpoint, and F = glutamine. The 13-run DSD above tests all six and returns a model in one block. A typical analysis of such a study isolates a handful of significant terms while leaving the rest at the noise floor.
Reading a DSD analysis
After running the 13 experiments and fitting the model by regression, a representative result for a CHO titer response (mg/L) looks like this:
- Temperature (A): coefficient −180 mg/L, p < 0.01 — strong linear effect, lower is better.
- Feed rate (D): coefficient +145 mg/L, p < 0.01 — strong linear effect.
- pH (B): coefficient +90 mg/L, p = 0.03 — moderate linear effect.
- pH² (B×B): coefficient −110 mg/L, p = 0.02 — curvature: titer peaks at an intermediate pH, not at an extreme.
- DO, glucose, glutamine (C, E, F): not significant (p > 0.2) — inert across the tested range.
Interpretation: three factors drive titer, and the significant pH² term means there is an interior optimum for pH — something a two-level screen would have completely missed. The DSD has both screened (dropping DO, glucose, glutamine) and started optimizing (locating the pH peak) in 13 runs. A short confirmation run at the predicted setpoint closes the loop.
This is the headline benefit for biology: where bioreactor time is the binding constraint, a DSD reaches an actionable model in roughly half the runs of the two-stage alternative. It pairs naturally with the broader screening logic in our DOE for bioprocess optimization guide.
A free alternative to JMP's DSD
JMP popularized the definitive screening design and made it a one-click feature, which is why many people search for "jmp definitive screening design biological process optimization." But the DSD is a published statistical method, not a proprietary JMP invention — the construction is the Jones-Nachtsheim conference-matrix fold-over described above, and any tool can generate it.
A free browser-based DOE generator builds the same conference-matrix DSD, assigns your factors, randomizes the run order, and exports the run sheet — with no licence and no install. For teams that need an occasional DSD without a JMP seat, or for students and small labs, this is a practical free alternative for definitive screening work. You give up JMP's polished interactive analysis, but the design itself — the run sheet you take to the bench — is identical.
| Design | Levels | Runs (k=6) | Curvature? | Mains clear of 2FI? |
|---|---|---|---|---|
| Definitive screening (DSD) | 3 | 13 | Yes | Yes |
| Plackett-Burman | 2 | 12 | No | No (res III) |
| Fractional factorial 2⁷⁻² | 2 | 16 | No | Some (res IV) |
| Full factorial 2⁶ | 2 | 64 | No | Yes |
| Central composite | 5 | ~50 | Yes | Yes |
Build a DSD free
You do not need a JMP licence to run a definitive screening design. A free design of experiments calculator builds the conference-matrix DSD for 4 to 8 continuous factors, sets the three coded levels, randomizes the run order to guard against time trends, and exports a clean run sheet for the bench.
Enter your factor names and their low, center, and high values, and the tool returns the 2k+1-run design ready to execute. After the runs, fit the model to rank the significant main effects and curvatures. It works alongside the other designs in the same generator, so you can screen with a DSD and, if interactions turn out to dominate, augment toward a response surface without switching tools.
Generate your definitive screening design now
A free, no-install DOE generator: conference-matrix DSD, three coded levels, randomized run order, and run-sheet export.
Frequently Asked Questions
What is a definitive screening design?
A definitive screening design (DSD) is a three-level design, introduced by Jones and Nachtsheim in 2011, that estimates the main effects of continuous factors clear of two-factor interactions and can also estimate quadratic (curvature) effects. It does this in about 2k+1 runs for k factors — for example, 13 runs for 6 factors — which lets it bridge screening and optimization in a single efficient experiment.
How many runs does a definitive screening design need?
A definitive screening design needs about 2k+1 runs for k continuous factors: 11 runs for 5 factors, 13 for 6, and 17 for 8. The design is built from a conference matrix folded over itself plus one center run. Some implementations add a couple of extra rows to improve power, but the run count stays close to 2k+1 — far fewer than a screening design followed by a separate response-surface design.
How is a DSD different from a Plackett-Burman or fractional factorial?
Plackett-Burman and resolution III fractional factorials are two-level designs that alias main effects with two-factor interactions and cannot detect curvature. A definitive screening design uses three levels, keeps main effects completely clear of two-factor interactions, and can estimate quadratic effects. That means a DSD can sometimes identify the optimum directly, whereas a two-level screen must always be followed by a separate optimization design.
Can a definitive screening design detect curvature?
Yes. Because each factor is run at three levels (low, center, high), a definitive screening design can estimate quadratic terms and detect curvature in the response. This is its key advantage over two-level screening designs, and it is what allows a DSD to find a peak or optimum rather than just rank factors by linear effect.
Is there a free alternative to JMP's definitive screening design?
Yes. JMP popularized the definitive screening design, but the design itself is a published method, not proprietary software. A free browser DOE generator can build the same conference-matrix-based DSD, randomize the run order, and export the run sheet at no cost — a practical free alternative to JMP for definitive screening design DOE bioprocess optimization.
When should I use a DSD?
Use a definitive screening design when you have 4 to 8 continuous factors, you suspect curvature, and you want to screen and begin optimizing in one efficient experiment. It is especially well suited to biological process optimization, where runs are expensive and you cannot afford separate screening and response-surface studies. Avoid DSDs when you have mostly categorical factors or fewer than four factors.
Related Tools
- DOE Experiment Generator — Build definitive screening, Plackett-Burman, and response-surface designs free in the browser with randomized run order.
- CHO Process Troubleshooter — Diagnose the temperature, pH, DO, and feed factors you would screen in a bioprocess DSD.
- Seed Train Planner — Plan the expansion that feeds the bioreactor runs in your screening campaign.
References
- Jones, B. & Nachtsheim, C.J. (2011). A class of three-level designs for definitive screening in the presence of second-order effects. Journal of Quality Technology, 43(1), 1–15. DOI: 10.1080/00224065.2011.11917841
- Bai, Y., Wang, Y., Chen, X., Zhou, J. & Zhou, W. (2024). Enhancing early-stage cell culture process development efficiency using an integrated approach of high-throughput miniaturized bioreactors and definitive screening design. Biochemical Engineering Journal, 205, 109217. DOI: 10.1016/j.bej.2024.109217
- Mandenius, C.F. & Brundin, A. (2008). Bioprocess optimization using design-of-experiments methodology. Biotechnology Progress, 24(6), 1191–1203. DOI: 10.1002/btpr.67