When (not to) pre-build a STAR index

Question

[jhanks1981]

I am soon to lose my institutional HPC access, and want to port some of my projects to run on a commercial cloud. To avoid excessive costs, I only want to pay when I am using it, and not pay for any more resources than I need. Ideally, I would rent a high memory system for a short period of time in order to build genome indexes for STAR (which requires 64GB of memory, far more than any of my subsequent jobs should need). I also might need to provide other people with pre-built virtual machines that already have genome indexes built on them.

However, it seems that people recommend against using pre-built indexes. Or at least it was the overwhelming consensus in this thread: Pre made STAR Index?

So my question is: How I know when I need to build a new index? Every time I create a new installation? When I change versions of STAR? What kinds of problems can arise from pre-building a genome index?

Answer 1

[Thibault D.]

AFAIK, unless it is clearly written by developers, I build one index for a genome version/patch, for one version of a software (either majors or minors).

There is not software I know deep enough to be 100% sure the last commit did not change the index, even by a single maybe-very-important octet.

For STAR, 2.7.xxx, if you're upgrading from 2.6.xxx, then it requires new index build, and release notes are clear about that:

Important: the genome index has to be re-generated with the latest 2.7.0x release.

Authors did not mention new index changes with the last version. It's likely to be unnecessary, but I would re-build index.