Not an easy task by any means. You'll need experiments where your sample types of interest are the same, i.e. control and wt must be the same. You'll then need to normalise each experiment, and convert identifiers to a common type, such as nuID. nuID is the most reliable as it ensures that you're using the same capture sequences. Take each of your normalised matrices, and column bind them, same with the phenotype data. To find differential expression, you include your "batch effect", or cross experiment effect in your linear model design. I'd advise against combat, there was a paper released recently to show it inflated the effects that you wanted to see (paper here)
You have to remember that this is a very difficult task, and there are a lot of technical effects you're trying to overcome. Consider a non-parametric approach to this problem too, it may act as a validation of sorts. Perform each differential expression test in each experiment, then look for common differentially expressed genes across experiments.