Scientific Explanation and Intelligibility or Understanding. I believe the former is grounded in the Metaphysics of Causation while the latter is grounded in Epistemology.
You have good intuitions, but it's a little more complicated than that. For instance, the sciences are certainly concerned with argumentation forms such as abduction, explanation, and scientific induction. These forms have a property in common known as defeasibility. Since, you have an AI tag, let me try to explain this to you in the language of machine learning.
According to Russell and Norvig, supervised learning is generally described as taking a hypothesis space H of hypotheses each h of and determining some function f' that maps causal pairs of (x,f(x)) which are data each s in the data set S such that f' approximates f presumed to exist. (If you're a scientific realist!) Theories are thus built by determining good probabilistic and deterministic explanations (f') from H that are presumed to represent the external state of affairs (f). In other words build theories from hypotheses that correspond to reality, thus making theories true as well as using true theories to decide if propositions are true. It's cyclical. There's a lot going on there, so let's unpack a bit.
In this AI instantiation of the problem of induction, one has an analog between causation and deterministic and probabilistic mappings between domain and codomain values, so yes, causality and explanation are linked because theories play the role of f' and seek to explain how change occurs reliably, predictability, etc. But, the question of what constitutes a good f' and how to find it are controversial and often contextual by subject matter hence leading to the demarcation problem of science. The act of the induction itself IS epistemological because the algorithm by which f' is determined to approximate f from h is the analog of epistemic attitude. Why should h1 be better than h2 is value-laden! Some might claim that the h1 and h2 are equally sufficient and no choice can be made. This is the idea of the underdetermination of theory. The question of what constitutes x is Quine's ontological commitment. What are the best x's and how do they relate to each other is controversial. And how BEST to approximate f is value-laden based upon, in machine learning, the metric of performance often tied to quality-quantity-speed issues inherent in theory of computation. This is pragmatism at its finest!
So, to the point. The h candidate for f' to represent f is the same thing as saying competing hypotheses vie to be the theory that represents reality, and that the causal relations and properties between ontological primitives which include both attributes and relationships change from past to present (and eventually to future) in the same way that x's in S relate to each other by set-theoretic definition such that at t1 they map to f(x)'s (reinterpreted as x's) in S at t2 so one can predict f(x)'s at t3. It's a gestalt! One cannot have functions independent of definition and mappings and elements and sets anymore than one can disentangle explanation, syntax, semantics, causation, and truth. They're all interrelated, and not structured towards each other in neat parallels.
In other words, causation, explanation, justification, and understanding are all interrelated aspects of epistemology. A theory must be coherent and meaningful and explain the relationships between definitions and observations and to determine which propositions of the theory are valid, they must be compared to that which they correspond to, be examined for coherence among themselves, and measured pragmatically in all theories. This is the Quinian doctrine of holism and touches upon notions like atomicity and molecularity of propositions.
My recommendation to a good start would be to get a copy of Blackwell's Companion to the Philosophy of Science and do a lot of reading in the SEP to explore the nature of science and bone up on what exactly is epistemological, ontological, and axiological in character.
