A cross-sectional design is totally inappropriate for this type of question. The results are almost certainly driven by selection bias: i.e. depressed people behave differently.

Without seeing changes from before and after the dietary change, you basically can’t learn anything.

TLDR; nutrition research is almost always done poorly.