What would you check before trusting a dose-response curve from a screen?

Why

A fitted IC50 is only meaningful if the data constrain it. Without a plateau at the top, the maximal effect is extrapolated and the potency is unreliable. With only one or two points on the slope, small errors move the fit a long way.

A Hill slope well above 1 suggests aggregation, precipitation or cooperative artifacts rather than simple binding, and Feng and colleagues built screens for exactly this behavior. A slope below 1 can mean a mixture or partial activity. Precipitation at the top concentration flattens a curve and produces a false maximum.

Eastwood and colleagues' minimum significant ratio matters here too, since it tells you how large a potency difference the assay can actually resolve, and a difference inside that ratio should not be reported as a difference. Compound integrity is part of the check. Libraries contain samples that are the wrong compound, degraded, or at the wrong concentration, so confirming identity and purity by mass spectrometry for anything important is routine in careful groups.

Finally the curve should be reproduced from an independent powder sample and dilution series, because a single dilution error propagates through every point.