Skip to content

v1.4.2

Choose a tag to compare

@josruirod josruirod released this 02 Oct 10:09
· 66 commits to main since this release
11a0ffa
  • When processing large assemblies/contigs, MegaBLAST execution in step 2 may be too computationally expensive. Made the parallel processing more robust and efficient (so the largest contigs are processed first and would fail first but not impede the processing of the rest).
  • New parameter to let the user decide whether to parallelize and how the processing of the contigs that failed
  • Captured the situation and exit ILRA when contigs are just too large for the current RAM available