Response Surface Methodology, Explained
Part of our complete guide to DOE for biologists.
Once you know which variables matter and want the exact best settings — not just "higher is better," but where the peak actually sits — you've moved past screening and characterization into Response Surface Methodology (RSM). See Screening Design vs. Response Surface: Which Do You Need? if you're not sure you're at this stage yet; this article assumes you are and covers how the optimization design itself is actually built.
Why this needs more than a screening design
A screening design tests each factor at two levels, which is enough to tell you the direction that helps but not enough to tell you where a curve bends. RSM adds a third level and enough additional runs to fit a full curved model — technically a second-order polynomial — across your factors. That model is what lets you solve for the actual optimum, or balance two competing outcomes against each other, like maximizing yield while keeping an impurity below a threshold.
The central composite design
The standard way to build this is a central composite design (CCD): a factorial core (your familiar low/high combinations), a set of axial points that extend along each factor's axis to let the model detect curvature, and center points repeated to estimate pure experimental error. How far the axial points extend is a real design choice, and it's the one worth understanding before you run one.
Why BioDOE uses a face-centered design
Some CCD variants place axial points outside your original low/high range, which means the design is technically asking you to test a setting you never actually specified as within bounds. BioDOE uses a face-centered CCD instead, where axial points sit exactly at the edges of the range you already defined — every run stays inside the box you set, nothing extrapolates beyond a temperature, concentration, or time you told BioDOE was a real, testable value. The trade-off is a small amount of estimation efficiency compared to some other axial placements; the benefit is that you never have to double back and ask "wait, is that setting even physically achievable?"
What it doesn't do
RSM tells you the shape of the response inside the range you tested — it doesn't tell you what happens outside that range, and it isn't a substitute for screening first. Running a full response-surface design on a long list of untested variables wastes runs modeling curvature for factors that might not even matter; screen first, then bring only the variables that survived into this step. See Which DOE Design Should I Use? for that full decision path.
Optimization (response surface) designs are a Pro feature in BioDOE; screening and characterization designs are available on the free Starter plan.