PROKKA detects fewer than half the features, compared with Prodigal alone

[willnotburn]

I ran PROKKA and Prodigal (in an independent run) to detect CDS. The resulting gff file generated by PROKKA contains fewer than half the features than the gff file from Prodigal. I also manually checked out some contigs to confirm fewer features detected by PROKKA. The PROKKA log shows it used the same version of Prodigal I ran independently.

PROKKA: 5313886 CDS detected Prodigal: 12330889 CDS detected

I am probably not understanding enough of under-the-hood of PROKKA. Does it impose additional quality filters to the ones default in Prodigal? Also (may be important): while I ran Prodigal on the entire assembly, PROKKA would take too long, so I split up the assembly into about 80 smaller parts, ran PROKKA on each split part separately, and then concatenated the results.

[spock]

AFAIR, prokka by default excludes edge run-off CDs when calling prodigal. More importantly, prokka expects a single genome, and not a metagenome (unless metagenome mode was added in the last year or so?), and calls prodigal in "single" mode. Finally, and insignificantly, prokka will drop CDS overlapping xRNA.

With a few simple adjustments you should be able to get exactly the same results as from prodigal alone.

[tseemann]

Thank you @spock for the excellent answer.

Prodigal in Prokka is using different options to you. Try --metagenome option.

  --metagenome       Improve gene predictions for highly fragmented genomes (default OFF)

[felipelira]

So, considering that PROKKA has the line:

my $cmd = "prodigal -i \Q$outdir/$prefix.fna\E -c -m -g $gcode -p $prodigal_mode -f sco -q";

Change it just removing '-c' should give the same number of genes in comparison with the PRODIGAL output?

[tseemann]

I guess so, yes. Assuming same version of prodigal and same -g, -m, -p modes.

[thierryjanssens]

The --metagenome option does not result in the same behaviour as leaving out -c in the standalone prodigal version (tested with --kingdom Viruses). Is there an option to include this as a new feature in the next version of prokka? This would be instrumental in the annotation of scaffold from a fragmented metagenome.