Preclinical testing takes the candidate molecule and studies it in cells and in animals. The aim is not to prove the drug works in people, which cannot be done yet, but to gather enough evidence that exposing humans is defensible. Two things are checked. First, safety: does the molecule damage the liver, heart, or other organs, and at what dose does harm appear? Second, behavior in a living system: is the molecule absorbed, does it reach the intended tissue, and does it stay in the body long enough to matter?
Most candidates fail here. A molecule that looked excellent in a computer model can turn out to be toxic, to be broken down too quickly, or to have no measurable effect in a living animal. Because this stage is comparatively cheap and involves no human risk, it is designed to catch those failures early, before the far more expensive human trials begin.
Suppose a designed molecule binds its target beautifully in a simulation. In animal testing it might be cleared from the bloodstream within minutes, so it never reaches the target tissue at a useful concentration. Nothing is wrong with the design logic; the molecule simply does not survive contact with a living body. That candidate is dropped, and the next one is tested.