It is not easy to give a simple answer to the question of how exhaustive analysis should be. As a rule of thumb, we would suggest that there should be enough information to convince you (or another expert) as to the decision that you have made.
So for cases where both predictions are concurring and there is little concern in that call from your own assessment, it should not be necessary to add anything. If predictions disagree and you take a conservative approach, then it is doubtful that a regulator would take much concern as you are not increasing the risk of exposing a potential mutagen to the human population, since testing or controlling will be the consequence of such a call.
If, however, you over-turn a positive prediction, then the level of analysis will need to be much higher. This could be a discussion highlighting sound scientific reasons for challenging the call, such as irrelevant training examples, poor learning from a statistical model, a lack of more appropriate examples which are supplied etc.
Our collaborative publication gives more examples and can be found here.