petercombs
commented
5 years ago Okay, so one problem with using the NCBI server to do blast is that it fails on relatively large results, ironically exactly like the ones we are getting. There may be some way to filter this, but it's not obvious how to do that in an easy programmatic way. We don't know in advance how far up the tree we'll need to go to get an outgroup, for instance.
So probably the long-term value here is in making a complete package that will do the selection calculations on even more MPRA data, with the operation on the Kircher data being a nice proof-of-concept. To that end, I should:
Include the aligners in the conda installation (where applicable?) to minimize the number of pre-installation steps that need to be done.
Give the option for blast to be done on the NCBI server. To limit the amount of requests, one could imagine having a maximum number of submissions, and above that requiring a local download, but I would think that this is not actually a big deal, since the refseq_genomes is pretty big, so you'd have to do a lot of blast searches to make up the difference. I'm still glad I have it locally, but no sense requiring it.