Improving performance and question about the output files from the specificity command

[charlesfoster]

Hello,

Thank you for this tool.

I generated a tiling scheme using:

• an input MSA with hundreds of genomes from NCBI for my virus of interest
• a BLAST database of non-specific sequences with many genomes of common circulating viruses + the human genome

In order, as directed in the README.md I ran 1) olivar build, 2) olivar tiling, 3) olivar specificity. The output naively looked good, i.e. ~96% coverage of my MSA of interest and the olivar-specificity_pool-1_ns-amp.csv and olivar-specificity_pool-2_ns-amp.csv files were empty. Accordingly, we ordered the primers to test on some clinical cases, but the results weren't good, e.g. a maximum of ~45% coverage even on samples with a very high viral load (as confirmed with qCR) and amplification of human sequences.

I'm now trying to figure out why the primers didn't work well in the real world as opposed to in silico. One thing I noticed was that despite the *_ns-amp.csv files being empty, the olivar-specificity_pool-*.csv files have some to many non-zero values in the 'BLAST_hits' column.

Q) Is this a red flag or is it normal?
Q) Am I correct that these represent blast hits against the non-specific sequences, or are they actually hits against the target MSA? If the latter, is it instead bad to see a value of zero for some rows?

Otherwise, do you have any other general tips for a potential re-design? For example, might I have used too many input sequences in the MSA? I thought it seemed like it still worked OK because of the good 96% coverage of the reference MSA in the logfile.

Exact commands used:

makeblastdb -in non_target.fasta -dbtype nucl -title non_target -parse_seqids -hash_index -out non_target  -logfile makeblastdb.out
mafft --auto --preservecase --reorder --adjustdirection --thread 8 target_virus.no_ambig.fasta > target_virus.no_ambig.aligned.fasta # followed by manual verification
olivar build --msa target_virus.no_ambig.aligned.fasta --db non_target --output olivar_target --title olivar_target --threads 12
olivar tiling olivar_target/olivar_target.olvr -o olivar_target --max-amp-len 1200 --min-amp-len 800 --check-var --seed 10 -p 12
olivar specificity olivar_target/olivar-design.csv --max-amp-len 1500 --pool 1 -d non_target -o olivar_target/ -p12
olivar specificity olivar_target/olivar-design.csv --max-amp-len 1500 --pool 2 -d non_target -o olivar_target/ -p12

Thanks

[mxwang66]

Hi @charlesfoster,

Sorry for the late reply, and I'm sorry the design didn't work for you. Seems like your issue is low coverage on real samples. If that is the case, maybe it is not because of primers being non-specific, but not sensitive enough (e.g., mismatches between the primers and the DNA templates in the real samples).

Did you check the output html file with primers and SNPs (sequence variations). If you see a lot of variations, it might indicate more mismatches between primers and templates. A solid way to validate is to use a viral standard that exactly matches the primers. Or do you have evidence that the low coverage is due to non-specific?

Best,
Michael

[charlesfoster]

Hi,

Thanks for the reply.

We resequenced the samples after redoing the amplification with different PCR conditions. After this optimisation we received much better (>90% complete) genomes, which is a nice improvement. However, there is still room for improvement so we will keep tweaking the design and conditions.

I check the specificity of the primers against one of the near-complete genomes we recovered. Most match well with no mismatches; only a few primer sequences have one mismatch. The only reason I considered non-specificity is because in the first sequencing run there were many human and other bacterial reads rather than our virus of interest.

Could I just follow up on the specific questions from the OP?

Q) Is it a red flag to have some to many non-zero values in the 'BLAST_hits' column of the olivar-specificity_pool-*.csv files?
Q) Am I correct that these represent blast hits against the non-specific sequences, or are they actually hits against the target MSA? If the latter, is it instead bad to see a value of zero for some rows?

Thanks

[mxwang66]

Happy to hear that you had better results!

For BLAST_hits in olivar-specificity_pool-*.csv, the numbers are BLAST hits of individual primers (against the non-specific db), so they don't necessarily mean non-specific amplification. I usually use them as an estimate as how "bad" is this primer in terms of specificity. Those in ns-amp and ns-pair are actual predicted non-specific amplicons, but in practice non-specific could be very messy and hard to predict.

Lmk if you run into other issues or have any other questions!

Michael