Plate layout and controls
Most assay failures are not chemistry. They are layout: edge effects mistaken for biology, a control in one corner, a candidate order that matches the pipetting order, and a comparison between plates that were run on different days with different reagent lots.
Edges are different
Outer wells evaporate faster, sit at different temperatures and can receive different amounts of light. Over a long incubation, the edge of a plate can drift far enough to change a result. Three habits handle it: fill the outer ring with buffer and use the interior for samples, keep plates in a humidified incubator, and check for a position effect by plotting the signal by row and column before interpreting anything.
Randomize, or at least rotate
If candidates are laid out in the same order on every plate, any position effect becomes a property of the candidate. Randomizing positions removes the correlation. Where full randomization is impractical, rotating the layout between replicate plates achieves most of the benefit and keeps pipetting sane.
The same applies to time. If one candidate is always processed first, its incubation is always shortest.
Controls that belong on every plate
- A reference molecule, which makes plates comparable and makes drift visible.
- A negative control matched in format, not just buffer.
- A blank, to define background.
- A saturating positive, to define the top of the window.
Controls belong distributed across the plate rather than in one block, because a block of controls in one corner measures that corner.
Judging whether the assay is good enough
The signal window and its variability decide whether a screen can distinguish anything. A separation statistic such as Z prime, computed from the positive and negative controls, is the standard summary for a screening assay, and it is worth computing per plate rather than once during development.
An assay with a narrow window can still be useful for large effects and cannot support fine ranking. Knowing which situation you are in before running the panel saves arguing about small differences later.
Replicates that mean something
Replicate wells on one plate measure pipetting. Independent experiments on different days, with fresh dilutions, measure the thing you actually care about. Report the spread across independent experiments, and treat differences smaller than that spread as unresolved.
For a panel, running everything on one plate is the cleanest comparison, and when a panel is too big for one plate, use a common set of bridging controls on every plate and normalize to them.
Recording it
Keep the layout with the data, machine readable. A plate map in a spreadsheet tab is enough. Results that cannot be traced back to a well and a lot cannot be checked later, and the checks are what makes a ranking defensible.