Lisa Feldman-Barrett's Theory of Constructed Emotion (TCE) makes good use of predictive processing and is relevant to these topics.
TCE starts by dispelling the notion that there are clear neural "fingerprints" associated with any emotion. We imagine there are fear or anger brain regions, and that's not accurate. (This may be taking it slightly too far. e.g. I'm pretty sure the amygdala is involved in every instance of what we call fear, but the point largely stands - like with love, there's too much variance to identify neural fingerprints for any category; each instance of emotion shows unique network activity.)
Instead, each emotional instance is a concept-driven top-down, embodied simulation of how we expect the body to feel about a given situation. Instead of emotions happening passively, the brain predicts then selects for emotional reactions.
For the brain, the body is an environment that needs to be interpreted and managed. The way our body feels on the inside is called interoception. Interoception refers to things like pulse, oxygenation, hunger, thirst, tiredness, signals from your organs, etc. Prior to a meditation practice, I was much less aware of interoceptive nuances. (I was also allergic to the term "energy" as related to chi or reiki or whatever. Now when I hear "energy healing," I think "interoceptive sensations work" and it sounds much better.)
We already know that interoception is subject to conceptual interpretation. E.g. Shachter & Singer's classic experiment showed that participants who were primed for fear on a suspension bridge were much more likely to interpret their bodily sensations as attraction and hit on the cute research assistant than were participants on a stable bridge. TCE explains how this works.
TCE defines the Interoceptive Network, which relates to DMN and Salience Network, and which regulated "body budget," allocating bodily resources to the demands of the situation. When body budget is in equilibrium, it means positive affect - the body feels good. When it's out of balance, then there is negative affect - the body feels bad. These are protoemotions, already felt in the protoself in infancy. It's why babies cry despite not yet having a "self" who feels upset.
Emotions are top-down filters that interpret interoception. Filters are influenced by: 1. personal priors and 2. social concepts about emotions. This second point is interesting. Apparently, Tahitians have no concept of "sadness," instead saying they feel ill or low-energy. Conversely, learning about other cultures' emotions can expand our repertoire of emotional experiences. It's not that you never feel these prior to learning of them, but that you may not categorize them as such.
TCE is important because it describes how concepts and interoceptive affect (respectively on the narrative and protoself levels) combine to create experience on the core self level. Also because it illustrates how top-down simulations filter sensory data it deems relevant. And because it describes the network nature of experiences: brain activity related to sensation, perception, memory, cognition, etc. all connect. It's a good model to know for understanding hyperpriors.