Skip to content

0.9.2

Choose a tag to compare

View all tags
@jdcla jdcla released this 14 Apr 17:09
· 73 commits to main since this release
0.9.2
cf5dcc0

📢 This release contains several version-breaking changes so read the changelog carefully if working with old databases.

h5 database

  • Optimized application of the polars framework and removal of pandas to load the GTF file properties, speeding up parsing time ~10-fold.
  • No more hardcoded loading of GTF columns unless a required property. This makes RiboTIE more compatible with GTF's in general. The predicted ORF table will therefore have different column names for some of the transcript properties that were previously hardcoded. e.g.:
    • support_lvl -> transcript_support_level
    • biotype -> transcript_biotype
  • Implemented LazyFrame loading for ribosome profiling data and optimized some code fragments, speeding up ~4-fold and lowering memory requirements.
  • Vectorized protein sequence storing—reducing storage space significantly (to 10% for GRCh38 version 110)

Data Loading

  • Before, only translation initiation site were annotated if a start_codon feature was present in the GTF file. Now, all first nucleotide sites regions annotated as CDS in the gtf file are saved as translation initiation sites. This improves compatibility with custom gtf files. To differentiate between CDS accompanied by a start codon and stop codon we included the properties has_annotated_start_codon and has_annotated_stop_codon within the hdf5 db.

Post-processing

  • New strategy to filter out ORF duplicates (sharing genomic coordinates) from the final list. See Filtering in RiboTIE README.md!
  • Completely re-wrote post-processing by polars'izing the code, Increasing post-processing times ~5-fold.
  • New post-processing attributes such as read_coverage, read_entropy, etc.
  • Lowered cutoff for positive set to 0.125 as many obvious false positives are now filtered out.

Other

  • Combine samples using group names, outputting results by their key values (With help of @NicolasProvencher)
  • Some refactoring and bug fixes.

Known issues: loading BAMs can files be slow and cause OOM. High RAM requirements can be >100Gb, but is highly dependent on individual samples.

Contributors

NicolasProvencher
Assets 2
Loading