- Flanking sequences are not called correctly
- Causes spacers to be called as flankers
- Speed degrades rapidly with the number of patterns identified
- Probable cause is the singleton finder
- In some circumstances closely related groups are not getting split apart
- The final output coverage of the spacer does not always correspond to the number of sources
- Spacers go missing in the graph from Crass, but they are present if you perform a regular overlap assembly
- This is not a sequencing error issue as even when searching with mismatches, there is not sign of the spacers
- Spacer strings only get created during the initial pass of the graph building. However, spacer objects can also be made from the p-graph
- This means that spacers may exist in the graph, but not in the output sequences. Maybe the cause of the above issue
- If a slave DR is a perfect palindrome of the master DR the offset will be undefined
- Multi-threading the search algorithms
- Using paired read information in the search algorithms
- Ability to run Crass on genomes or assembled contigs
- Output a gff3 formatted file
- Screen for known eukaryotic microsatellites for datasets that may be host contaminated
- Use read information better when building the graph
- Allow for multiple DR types to be in a single sequence. Very important if Crass is to work on genomes