I'm fairly good at finding bugs and understanding code. Then recently I've had to mentor a few junior engineers and it seemed they struggled a lot.

It's a mix of search space reduction + heuristics. You start looking for most probable areas of fault. This includes new commits. Then you try to partition the places where the bug could be by probablistic reasoning, logging as you go.

If these approaches don't work then you start questioning your assumptions. For example, if there is eatApple() method, what did you assume about this method?

Did you assume all apples are rex, or is the functionality actually eatFruit()?