REST doesn't care what spelling conventions you use for your resource identifiers.
HTTP expects that, to patch a resource, the PATCH request will target the same resource identifier as a GET request. (This is a consequence of how caching works, see RFC 9111.)
So if you are fetching the current representation of a "comments resource" via
GET /comments/1
The you (RFC 2119) SHOULD be sending patches to the representation of that resource via
PATCH /comments/1
so that general purpose caches can understand the semantics of what is going on, and do intelligent things.
An important idea in REST is the uniform interface constraint; among other things, it implies that everyone should understand the semantics of requests and responses the same way -- no one resource should be a "special case" that assumes different meaning to the messages.
It follows that "nested resources" should understand request messages the same way that everything else does; GET means GET, PATCH means PATCH, and so on.
As for the design of your resource identifiers, the machines don't care (that's part of the point of "uniform" interface), so choose spelling conventions that best meet the needs of the human beings you value.