Its a great question , but considering the abilities of network biology and systems biology tools, the central hubs (based on higest centrality) would provide you potential anawers. But you have have to be really cautious and must always make mannual judgement based on the overlap (with host / within the target organism) as well as look ar the score of deviation in expression , the less deviated targets would be relatively better choice

This question can have lots of answers, and all the answers can also be vague at the same time.

It all depends on how you define the term "suitable". If you are familiar with network centrality, then you will observe that there are numerous algorithm which identifies the most important nodes in a network. Being important, omitting them from the network will have a significant effect, and hence, the top nodes obtained from each of the algorithms justify themselves to be a potential target.

However, the story doesn't end here. As you mentioned, a protein to be called a target should affect certain pathways and biological processes. Also, in diseases like cancer, they should possess certain properties like patient survivability. Even after this step, factors like toxicity need to be addressed.

Unless you validate your findings experimentally, it is hard to tell that it is a suitable target. But yes, using a proper in silico methodology, you can identify the proteins which are "most likely to be called potential targets" in a disease.