On-target, off-target
Every effect a drug produces falls into two bins. An on-target effect comes from acting on the protein you designed it to hit — these include both the therapeutic benefit and any harm that flows from over-engaging the right target. An off-target effect comes from the drug binding *other* proteins it was never meant to touch. Most surprising side effects are off-target, and untangling which is which shapes the whole safety story.
Quantifying selectivity
Selectivity is how much a drug prefers its intended target over everything else. We make it a number by comparing potencies: the IC50 at an off-target divided by the IC50 at the intended target. If your compound has an IC50 of 2 nM on the target and 2000 nM on a related protein, you have a thousand-fold selectivity window. The bigger that gap, the more room you have to dose the drug into its useful range before the off-target wakes up.
Selectivity ratio = IC50(off-target) / IC50(on-target)
Compound A: on-target 2 nM, off-target 2000 nM
ratio = 2000 / 2 = 1000x (good selectivity)
Compound B: on-target 2 nM, off-target 6 nM
ratio = 6 / 2 = 3x (poor; off-target hit at similar doses)
Rule of thumb: you generally want the on-target effect to occur
well below the concentration where off-targets start to engage.The therapeutic window and index
Now zoom out from single targets to the whole patient. Below a certain exposure the drug does too little; above another exposure it starts causing harm. The gap between "enough to work" and "enough to hurt" is the therapeutic window. A drug lives or dies by how wide that window is. Warfarin and many cytotoxics have famously narrow windows — a small dosing error tips you from benefit into toxicity — while many common antibiotics have windows so wide that precise dosing barely matters.
The therapeutic index turns this into one ratio: a measure of the toxic dose divided by the effective dose. Classically it is the dose toxic to half the population over the dose effective in half the population. A therapeutic index of 2 is white-knuckle; an index of 100 lets a clinician dose generously. This single number, related closely to the safety margin you carry from animal studies into humans, is one of the first things a project asks about a candidate.