Unfortunately, @1111, I can't. 
estimating required ram depends on:
- data size
- data complexity
- other processes you have running
- probably other stuff I'm not thinking of.
Your data set isn't huge, so trial and error will probably get you there. Before you go buying more RAM, you should know that, though out-of-memory errors are common, sigkill errors could happen for other reasons too: maybe the system went to sleep, rebooted, ran out of storage space, or something.
Some things to try:
- make sure your machine doesn't suspend/sleep/hibernate while you're running the command. This could cause a sigkill
- shut down any extra processes when you run
denoise-paired. Firefox is great, but will take up memory you may need. - try adjusting
--p-n-reads-learnto a smaller number. Maybe 100,000? This parameter constrains the size of the data set used for training the error model, and running at a lower number should reduce memory required in exchange for some loss in model quality.
In a pinch, you could also rent a computer on the interwebs and run your analysis there. Setup would be a little hassle, but they can be quite cheap, and the time savings/convenience of freeing up your computer could make it worthwhile.