First, one potential caveat.
Above, you say "I have obtained several GO terms (indicating function and biological process) for several proteins". Does this mean you do, or do not have a genome-wide dataset? If you only have a handful of genes, it may not be possible to accurately define what is going on for pathways whose members are largely missing from your list. Also, are you referring to gene expression data, or proteomics?
What you are proposing is challenging because it is cell-type and cell-state specific. In other words, because gene expression varies according to cell type as well as current environmental conditions, there is not one connection hierarchy that describes how pathways relate, regulate, or influence each other - there is a spectrum of them.
To create a resource that did this, i.e., that defined connection hierarchies across a multitude of cell types and states would require high quality data from specific cell types that was well-annotated. At present, whether or not this resource already exists (and its quality, if it does exist) depends on what exactly you are studying. Even if you do find such a resource, the burden of proof is on that scientist to demonstrate why that pre-existing annotation scheme is thought accurate enough to use to annotate her or his data.
There are a few other general approaches to doing this, though.
1) If your sample size is large enough, you may be able to derive pathway relationships empirically using your own data. As an example for how this would work, consider 2 pathways, X and Y. Suppose across all 87 of your samples, in every case in which pathway X is upregulated, pathway Y is downregulated. Then, empirically, the data may be indicating that the pathways have feedback on one another (alternatively, it could mean something else is responsible for those effects).
2) One can go to the literature and manually curate the relationships between the pathways about which they are most interested. This has been done, including in high-impact publications, because often there is no sufficiently exact connection hierarchy to date. This may seem arbitrary. In other words, what keeps the scientist from cherry picking what pathways to include in their hierarchy? Having said that, keep in mind that in many cases the use of published pathway ontologies may be no less biased. Finally, many of the published gene sets themselves are actually curated in this way, biased or not. Consider signature set C2 from MSigDB.
Consider, for instance, the hallmark gene sets found in MSigDB. Note that, as the description on this page states, these gene sets are formed by overlapping gene signatures in many datasets in order to find pathways that are generally coherent across multiple experiments. However, this may mean that the unique biology found in one specific experiment isn't perfectly represented by the hallmark set.
Overall, these limitations mean that it is important to look at the dataset from which a given pathway was constructed in order to begin to understand its applicability. For instance, this pathway defines targets of TGFB in hepatocellular carcinoma cells. Because TGFB displays extreme variability in its range of effects depending upon the tissue in which it is found, use of this pathway may or may not be meaningful outside of hepatocellular carcinoma.
-- With all of that as background, here are some places to start --
1) Consider reading the WGCNA manuscript, an article with over 15,000 citations to date. This manuscript will discuss gene co-expression in a way I think that will be helpful to you, and if you identify the functionality you need from the paper (the introduction defines 10 or so analytical goals), you can try the WGCNA package.
2) It is possible with much less work to look at co-variance between lists of genes in your data if you have it narrowed down to a handful of pathways.
Hope some of this helps.