Let's just assume dose-response is linear. So, go find what concentration of chemical X (medicine, food additive, cosmetic, whatever) kills 50% of the bacteria/rodent/etc population. Draw a line from there down to 0% dead at 0 concentration.
Now decide what % of deaths is acceptable, given the benefit of chemical X. Use the line we just drew and that tells you what the regulations should set for max concentration under as-directed usage.
This ignores facts like
response curves are very often literally curves, sometimes with more than 1 minimum
response curves (and lines!) often depend heavily on all sorts of other chemicals' presence in the environment, in varying mixtures & amounts
It's a combinatorial explosion we can never meaningfully test against. We do what we can, with the tools that we have.
