Question about filtering external CDSs supplied with -c #83

Activity

1. alekseyzimin commented on Sep 10, 2026

   

   alekseyzimin

   on Sep 10, 2026

   Owner

   More actions

   Identity and similarity for external CDSs is by default 100%, as they match the genome exactly. Ordinary proteins from related species do not, there are differences in the translated sequence. I do not have any way to validate external CDSs except for matching them with transcripts. That is why the CDSs supplied with -c must be of high quality.
2. hibiki-kato commented on Sep 10, 2026

   

   hibiki-kato

   on Sep 10, 2026

   Author

   More actions

   Thanks, that makes sense. I think it would still be useful to have an option for external CDSs that are not necessarily trusted.
   Looking at the current main code, _EXTERNAL seems to bypass filtering for unmatched CDSs at two stages:

   1. In eviann.sh L769, _EXTERNAL bypasses the splice-site Markov/WAM threshold.
   2. Later, external candidates are excluded from the normal protein-specific ranking in filter_unused_proteins.pl L136 and filter_unused_proteins.pl L156, and are explicitly retained at filter_unused_proteins.pl L175.

   The second bypass makes sense to me given your explanation, since identity / similarity are not meaningful for an external CDS.
   But the first bypass seems independent of the missing protein-alignment metadata, since the splice-site Markov/WAM scores can still be computed for an external CDS.

   Would it make sense to add an option for less-trusted external CDSs where this first bypass is disabled, while keeping the current -c behavior for genuinely high-confidence CDSs?

   flowchart TD
       A["Unmatched protein/CDS candidates"]
       A --> B["Compute splice-site Markov/WAM scores"]
   
       B --> C{"_EXTERNAL?"}
       C -- Yes --> E["Downstream processing"]
       C -- No --> D{"Splice scores pass?"}
   
       D -- No --> X["Reject"]
       D -- Yes --> E
   
       E --> F{"_EXTERNAL?"}
       F -- Yes --> H["Remove redundant intron-chain models"]
       F -- No --> G{"Pass normal unmatched-protein filtering?<br/>readthrough / ORF validity<br/>identity / similarity<br/>protein support + locus ranking"}
   
       G -- No --> X
       G -- Yes --> H
   
       H --> I["Final set"]
   

   Loading