Okay, first, I would say to everyone planning a high-throughput experiment: Hypothesis first, then experiment. Otherwise you're setting up an expensive and time-consuming test that's not going to answer the questions you want answered.
In this case, the question your experiment is designed to answer is "What genes are crucial to growth in vivo that are not crucial to growth on a plate?", which is probably not the clearest model for vaccine/antibiotic design.
I guess you could say that a difference between the plate sample and the in vivo sample is the presence of a host, and therefore it's possible that upregulated pathways represent the bacteria's host adaptation, and maybe you can attack the bacteria by attacking its ability to colonize the host. A gene involved in immune evasion could make a good target.
(That's a very messy model, as there are tons of other differences going on, but since you've already run the experiment it's probably the best you've got.)
So then, use your data for a new round of hypothesis generation. Take your pathways and come up with a way that they might be allowing the bacteria to adapt to the host-- maybe a secretory pathway that could be interfering with the target immune system, or some membrane signalling molecules that get upregulated, or whatever.
Now you have a hypothesis, so design an experiment. For instance, take a key upregulated member of a pathway. The hypothesis is that it's crucial to growth in a host but not to growth in general. So, knock it down, either with chemical inhibitors or with genomic editing, and see if it's less effective at host colonization.
(Although host colonization effectiveness is pretty difficult to measure ethically if your target organism is humans. But you have some sort of proxy or model system to measure that, right?)