What makes a chemical probe good enough to draw biological conclusions from?

Why

Arrowsmith and colleagues set out working criteria used by the Structural Genomics Consortium for epigenetics probes: potency below 100 nanomolar on the target, more than 30-fold selectivity over related family members, profiling against a standard off-target panel, demonstrated on-target cellular activity below 1 micromolar, and an inactive close analog supplied with the probe. They also argue that one probe is not enough, and that a structurally distinct orthogonal probe reduces the risk of shared off-targets. Their warning is concrete.

Staurosporine was described as a protein kinase C inhibitor and turned out to hit many kinases. LY294002 was sold for years as a selective PI3 kinase inhibitor after it was known to inhibit many proteins at the concentrations used. Tens of thousands of papers each year rely on probes of unknown quality.

The practical reading for a scientist is to check a probe's current characterization rather than its original paper, to use the inactive analog in every experiment, to work at the lowest concentration that engages the target, and to confirm the key conclusion with an independent chemical series or a genetic perturbation.