High throughput screening is the practice of testing large compound libraries against a biological read-out in miniaturised, automated format. In antibacterial discovery it works, but the economics are unusually brutal, and the reason is worth understanding before designing a campaign.
The mechanics
- Plate formats. 96 well for careful work, 384 for most campaigns, 1536 where volumes and cost force it.
- Read-out. For antibacterials the primary read is growth inhibition, measured by optical density or by a metabolic indicator such as resazurin. Whole cell, phenotypic, no target required.
- Controls. Every plate carries growth and sterility controls plus a reference agent, because plate to plate variation is the main source of false calls.
- Confirmation. Primary hits go to a dose response, then to a proper MIC determination. A primary hit is a hypothesis, not a result.
Why antibacterial screens have such low hit rates
Target based screening dominated the 1990s and early 2000s and produced very few clinical antibacterials. The reason is now well documented: compounds that inhibit a purified essential enzyme frequently fail to reach that enzyme inside a living cell. The bacterial envelope, and especially the Gram-negative outer membrane, filters out most of medicinal chemistry's usual chemical space.
Whole cell screening avoids that failure mode by testing the real question directly, but it pays for it in two ways: the hit rate against Gram-negatives is very low, and a hit arrives with no mechanism attached. You then have to work out what it hits, which is its own project.
The chemical space problem
Compounds that accumulate in Gram-negative bacteria tend to be small, polar and to carry an ionisable amine, which is close to the opposite of a typical drug-like screening library optimised for oral absorption in humans. A conventional corporate library is therefore not a good antibacterial library, and screening more of it does not fix that. The Gram-negative bacteria page covers the accumulation rules in detail.
What a campaign actually costs
Costs vary enormously with format, compound supply and whether the work is internal or contracted, so treat the shape rather than the number as the point:
| Stage | What drives the cost | What triage changes |
|---|---|---|
| Primary screen | Plates, reagents, compound supply, instrument time | Fewer plates for the same decision |
| Hit confirmation | Dose response in triplicate | Fewer false leads reaching this stage |
| MIC panel | One plate per organism per compound | Panel scoped to organisms with a real chance |
| Mechanism work | Resistant mutant generation, sequencing, imaging | Starts with a hypothesis instead of nothing |
| Calendar time | Queueing, repeats, reagent lead times | Weeks removed at the front |
Triage before the plate
In silico triage does not replace a screen. It changes what goes into it. Three things are worth deciding before a library reaches a plate:
- Is the scaffold already known to fail? If the closest published analogs have no Gram-negative activity, the series will need a permeability strategy, not more plates.
- Which organisms belong in the panel? Running six organisms when the compound class has a realistic chance against two is a way to spend five sixths of the budget confirming what was predictable.
- Which analogs actually differ? A series often contains many compounds that will behave identically. Ranking removes the duplicates before they consume wells.
The output of that triage is a shorter plate list with a written reason for each inclusion. That is also what makes the result defensible in a program review, which matters as much as the wells saved.
Where imaging read-outs fit
A growth curve says whether a compound worked. Imaging based high content screening can say something about how, by reading the shape and organisation of the treated cells. It costs more per well and it answers the mechanism question earlier, which is the trade a program makes deliberately rather than by default.
Screen a compound now
The screen below runs the triage step on one compound against a six strain panel. It returns a predicted band per strain with the resistance mechanism named, which is the input to deciding what the real campaign should contain. Details of the method are on the how it works page.