Two kinds of assay, two kinds of question
An assay is a controlled measurement that turns a molecule's effect into a number. The two great families are biochemical and cell-based. A biochemical assay mixes purified protein with substrate and your compound in a tube or well, then reads out activity directly — for an enzyme, how much product forms. It is clean, fast, and tells you unambiguously that the compound acts on that protein.
A cell-based assay puts the compound on living cells and measures a downstream consequence — a reporter gene lighting up, a receptor signaling, cells dying or dividing. It answers a harder, more honest question: does the compound do the right thing inside a cell, where it must first cross a membrane and survive? That realism is the whole value, and also the whole difficulty.
Trade-offs that matter at scale
Biochemical assays are cheaper, faster, and easier to interpret, which is why classic HTS campaigns often start there. But purified protein can behave differently from the protein in a cell, and a biochemical hit may never reach its target through a membrane. Cell-based assays restore biological context but bring noise: cytotoxicity masquerading as activity, off-pathway effects, and far more variability. Each well is a tiny living system having a bad day or a good one.
The number you usually report from a confirmed hit is an IC50 — the concentration that halves the measured activity. An IC50 from a biochemical assay and an IC50 from a cell assay are not the same animal: the cell value folds in permeability, efflux, and protein binding. When the cell number is far worse than the biochemical one, that gap is itself a clue about what to fix.
The Z-factor: is this screen even runnable?
Before screening a million wells, you must know the assay can separate active from inactive reliably. The Z-factor captures this in one number. It compares the gap between your positive and negative controls to the scatter within them. A large clean gap with tight controls gives a Z-factor near 1; overlapping, noisy controls drag it toward 0 or below.
Z' = 1 − [ 3·(SD_pos + SD_neg) / |mean_pos − mean_neg| ] Z' > 0.5 excellent, large separation → fine for HTS 0 < Z' < 0.5 marginal, usable with caution Z' ≤ 0 unacceptable, controls overlap → fix the assay first Rule of thumb: never start a big screen below Z' = 0.5.