I have 100+ million paired-end reads from which adapter sequences have been removed but quality trimming has not yet been conducted. It seems most efficient from a memory utilization standpoint to conduct in silico normalization and trimmomatic steps FIRST and only once and then to use normalized/trimmed data for any subsequent trinity assembly trials. If so, which should come first, the normalization operation or the trimmomatic operation? Thanks!
Thanks for the advice and links, Brian. I've been considering BBNorm for all the reasons you mention in support of the software above and elsewhere.