Sometimes you can infer from evolution that a trait is useful. But only if the trait:

1. Is in a stable part of the genetic code, giving opportunity for bad traits to be weeded out.

2. Is not correlated with a different useful trait that overshadows its goodness/badness.

3. Shows up early enough to affect reproduction.

Human intelligence is quite recent, so prerequisite #1 is not met. We have no idea if prerequisite #2 is met, and prerequisite #3 is met.

So natural selection can't tell us if depression is/was useful. The kinks of human intelligence are not yet worked out.

#1 Dogs can get depressed so it doesn't need human intelligence. (https://en.wikipedia.org/wiki/Learned_helplessness)

#2 Depression has a major effect, I doubt it's overshadowed

#3 You get depressed teenagers

So I'd think 1,2 and 3 are met.

Re #1: There's absolutely no way to know if dogs experience "depression" in the same way humans do -- even assuming the outwardly symptoms are similar.

Re #2: Your doubt does not constitute evidence in any reasonable sense.

Evolution tends to keep things that increase survival chances, but it can also keep useless features. Also, when the environment changes, a previously useful feature may become useless. Evolution doesn't ensure that feature will disappear, unless it provides a major disadvantage.

By that logic, why can I get sick at all? Why must depression only have social roots?