A design space, and the experiments that find it.

Formulating a nanocarrier is a search problem. There are a dozen ways to make the particle, six or seven process factors that matter for whichever one you pick, and a response surface you cannot see. This walks the loop: narrow the technique from what the molecule is, bracket the factors, build a design that can actually estimate what you want to estimate, fit the surface to your results, and find the settings that satisfy every response at once. The designs are constructed from their definitions and checked against their own properties, not copied off a table. Everything you type stays in this browser, with one exception you press yourself: the PubChem lookup on the compound tab.

● your formulation data stays in this browser · the PubChem lookup is the one request, and only when you press it

Read the labels on the numbers. Two different kinds of thing live on this page. The experimental designs and the statistics are mathematics: they are exact, and they are tested in front of you at the bottom. The technique ranking and the starting factor ranges are formulation heuristics: they encode how an experienced formulator would open the problem, they are not predictions, and they are not derived from your compound by any physical law. Every block says which it is. Widen or narrow any range before you run anything.

Compound in, protocol out.

the loop, in four steps
01

Define the compound

Look the molecule up in PubChem by name or SMILES, or type the descriptors yourself. Molecular weight, LogP and TPSA come back calculated; dose and measured solubility are yours to supply, because no database holds your formulation's.

02

Select the technique

Nine manufacturing routes, scored against cargo, route of administration, target size and sterility. The rule that fired is printed beside every score, so you can disagree with it line by line.

03

Design the experiments

Bracket the factors the technique actually has, then build a design that can estimate what you intend to estimate: screening or response surface, with the resolution and the star distance derived rather than asserted.

04

Optimise and write it up

Fit the surface by least squares, check lack of fit against pure error, find the settings that satisfy every response at once by desirability, and read the protocol out at those settings.

What is in here

  • PubChem lookup for calculated molecular descriptors
  • Nine techniques, scored by itemised rules
  • Full and fractional factorial, Plackett-Burman, central composite, Box-Behnken
  • Least squares response surface with standard errors
  • Lack of fit tested against replicated pure error
  • Derringer and Suich desirability across every response
  • A protocol written at the settings you chose
  • Every construction asserted against its own definition on load

Research use only.

what this is, and what it is not

This is a teaching and planning instrument. It is not formulation advice, not regulatory advice, and not a prediction of what your process will do. The technique ranking and the opening factor ranges are heuristics: they encode how an experienced formulator would open the problem, and they are not derived from your compound by any physical law. Every computational result on this page has to be confirmed experimentally before it means anything. Nothing here has been reviewed by a regulator.

Everything you type stays in this browser. The one exception is the PubChem lookup, which is an explicit button: pressing it sends the text in the search box to the National Library of Medicine. Nothing else on the page makes a network request, and you never have to press it, because every field it fills can be typed by hand.

Compound resolution.

PubChem lookup, on request only

Pressing that button sends the text above, and nothing else, to pubchem.ncbi.nlm.nih.gov. It is the only outbound request this page can make. If you would rather not, fill the fields below by hand.

Retrieved properties
Identity
3D structure
no structure loaded

Drag to rotate.

The molecule.

calculated descriptors, entered by you

Needed for a real solubility classification.

Measured, not predicted.

Technique scoring.

a formulation heuristic, itemised so you can disagree with it

The factors.

a starting bracket, yours to change

A design of experiments does not need the range to be right. It needs the range to be wide enough to contain the answer and narrow enough to be runnable. If you know the real limits, type them in.

The design.

constructed from its definition, then checked against its own properties

Replicated, so pure error and lack of fit can be separated.

The surface.

least squares on coded factors

The settings.

Derringer and Suich desirability across every response at once

The protocol.

written at the settings you chose, deterministically

How the scoring works.

the rule classes, stated

A technique starts at zero and accumulates the weight of every rule whose condition its context satisfies. There is no model and no fitting: the ranking is the sum of a handful of signed weights, each of which is printed beside the score with the rule that produced it. Four kinds of condition exist: the cargo class, the route of administration, the target particle size relative to what the technique can reach, and whether the product must be sterile. A negative weight is a technique arguing against itself, not an exclusion, so a low-ranked technique is still selectable. The weights are formulation judgement. They are not fitted to any dataset, and they are not traceable to a citation, which is why they sit behind a heuristic label everywhere they appear.

What this does not do.

stated rather than discovered

Self-test.

the designs and the statistics, checked in front of you

A design of experiments is worth nothing if the design is wrong, and the failure is silent: you run the experiments, fit a model, and the model estimates something other than what you think it estimates. These assertions check the constructions against their defining properties on every load.

© 2026 Rizkin Labs ← back to the bench Legal · a teaching instrument, not formulation or regulatory advice