Every prediction falls into one of four boxes, and two of them are mistakes. A false positive is a drug the model flags as likely to fail that would in fact have succeeded. A false negative is a drug the model passes as low risk that goes on to fail. The names describe the prediction, not the drug: "positive" means the model predicted failure.
The trade-off is unavoidable because the two errors pull in opposite directions. Lower the threshold so that more candidates are flagged, and you catch more of the drugs that would truly fail — but you also discard more drugs that would have worked, so false positives rise. Raise the threshold, and you keep more candidates in play, but you let more eventual failures through, so false negatives rise. There is no setting that reduces both at once; you are choosing which mistake to make more often.
In drug development the two mistakes are not equally costly, and the cost is not purely financial. A false positive throws away a compound that might have helped patients — a loss that is hard to see, because the discarded drug is never tested. A false negative lets a doomed candidate consume years and large budgets before it fails in people. Which error a team should tolerate more depends on how many candidates it has, how expensive the next stage is, and how much it needs a success. That is a strategic judgment, not something the model can supply.