-
Notifications
You must be signed in to change notification settings - Fork 27
Zika virus annotation
- Explanation of the recommended options
- Nucleotide-level clade classification
- A note on the INDEFINITE_CLASSIFICATION_NN alert for Zika
GenBank is not currently using VADR to
automatically process Zika sequence submissions like it does for
SARS-CoV-2.
GenBank Zika sequence submissions are still all manually reviewed, but
curators may use VADR as part of that review, running v-annotate.pl on the
command line with options like those below.
Steps for using VADR for Zika annotation:
-
Download and install VADR v1.7.1 or later, following the instructions on this page. v1.7.1 is the first version that works with the Zika models; earlier versions cannot parse the model info file at all.
-
The Zika models are installed automatically by
vadr-install.shas of v1.7.1. To download them separately, the current package (version 1.7.1-2, gzipped tarball) is here. Unpack it (tar xfz <tarball.gz>) and note the directory it creates (<zika-models-dir>). -
WARNING: the
fasta-trim-terminal-ambigs.plscript will not exactly reproduce the trimming that the GenBank pipeline does in some rare cases, but should fix the large majority of the discrepancies you might see between local VADR results and GenBank results.To remove terminal ambiguous nucleotides from your sequence file
<input-fasta-file>and to remove short and long sequences to create a new trimmed file<trimmed-fasta-file>, execute:
$VADRSCRIPTSDIR/miniscripts/fasta-trim-terminal-ambigs.pl --minlen 50 --maxlen 12000 <input-fasta-file> > <trimmed-fasta-file>
- Run
v-annotate.plon the trimmed fasta file, with the options in the next section.
v-annotate.pl -r --r_file <zika-models-dir>/zika.rpn.fa --nosub \
--lowsim5ftr 15 --lowsim3ftr 15 \
--mkey zika --mdir <zika-models-dir> \
<trimmed-fasta-file> <output directory>
This matches the recommended usage in the model package's own 00README.txt.
| option | what it does |
|---|---|
-r --r_file <dir>/zika.rpn.fa |
seeds the alignment from the 8 model sequences in "replaced-N" form, which speeds up and stabilizes alignment of full-length genomes |
--nosub |
disables subgroup-based model selection; the package contains a single model, so there is nothing to select between |
--lowsim5ftr 15 --lowsim3ftr 15 |
raises the low-similarity thresholds at feature 5' and 3' ends, which suppresses spurious low-similarity alerts in the UTRs |
--mkey zika --mdir <dir> |
selects the zika model from the unpacked package |
The Zika package enables VADR's nearest-neighbor classification, so v-annotate.pl
additionally writes a .scn file assigning each sequence to an African, Asian or
American clade, together with the percent identity to its nearest and second-nearest
training sequences and the gap between them.
Expect this alert on nearly every Zika sequence. It is not an error and it is not fatal.
INDEFINITE_CLASSIFICATION_NN (nnindfcl) fires when the fractional identity gap
between a sequence's best and second-best clade falls below a threshold, 0.05 by
default. Zika's clades are far more compact than that default anticipates, so almost
every sequence trips it. On a 754-sequence benchmark spanning all three clades, 753
sequences (99.9%) were flagged. The median gap was 0.0049, and even the most divergent
clade in the set (African) had gaps of only 0.031 to 0.049.
The alert is never fatal. It does not change the annotation, the pass/fail status, or the clade assignment itself, and on that benchmark the clade assignments were 100% concordant with published assignments despite essentially every one carrying the alert.
If the alert is not useful for your application, the threshold can be lowered with
--nn_indefclass <x>. We do not recommend a specific value. The gap distribution for
Zika is narrow enough that the flag rate is very sensitive to the exact threshold chosen
(on the benchmark above, 0.005 flagged 54% of sequences while 0.002 flagged 2%), and
we have no dataset containing genuine Zika clade misclassifications against which a
tighter threshold could be shown not to miss real cases. The default reports a true
property of the data, which is that Zika's clades are separated by very little
divergence.
VADR v1.7.1's --draw_r2dt option renders the annotated secondary structure of each
sequence using R2DT. The Zika package ships the two
required R2DT templates (zika-linear, zika-circular) under r2dt-templates/, and
vadr-install.sh installs them so that --draw_r2dt works without any manual step.
Add --draw_r2dt to the command above to produce the drawings. See
r2dt-drawing.md
and r2dt-templates.md
for details.