I have RNA samples but with different read lengths (Eg, HiSeq 2x125 and NovaSeq 2x150bp data). I would like to do DE analysis on these samples. What do you recommend? Do you recommend to trim the reads to 125bp for the NovaSeq to compare with the other batches from HiSeq? Or any other suggestions.
You don't need to do anything, this is taken care of by the linear model - if you don't have any confounding effects.
If you do have confounding effects, e.g. all control samples on HiSeq and all treated on NovaSeq there's not a lot you can do anyway, other than be honest when you present your results.
Taken care of by the model means adding an extra 2 level factor for sequencer by sample, you need to do this anyway to take care of the batch effect between sequencer runs.
Yes, I would simply trim the longer one to the shorter read length, then map the data as usual. I doubt though that it makes a big difference on the global scale but the processing step is simple and it makes sure you have no mappability bias. If you do not have any other confounding effects then you can do a DE analysis without any extra covariates, I think it is always desirable to keep things as simple as possible and trimming is the simplest solution.