How to accelerate BLASTING

Question

[anikcropscience]

Hi, I have 10000 unassembled contigs from a metagenomic analysis. I have no idea which sequence belongs to which species. I ran Kraken classification but that is not enough as I still have half of the reads unclassified. I have collected the raw material for nanopore sequencing from an infected plant. I do not know what kind of pathogen (virus, fungus, or bacteria) caused it. So, I tried to blast each contig remotely using the following command:

blastn -query filtered.assembly.not.aligned.fasta -remote -db nr -out blastoutput_unassembled.txt -outfmt '6 qseqid sseqid evalue bitscore sgi sacc staxids sscinames scomnames stitle' -max_target_seqs 1 

But this process has been running for the last 7 days and still only results from 3000 contigs are available.

Could you please suggest if this process can be accelerated or any other alternative solution for the purpose?

[Joe]

Kraken (or I think Centrifuge?) would be the way to go IMO. They are specifically designed for this task, which BLAST isn't really. Check you are using the latest database versions etc. I could be wrong but I would expect them to be using datasets close to if not the same completeness as NR.

The most immediate answer to your question though is : don't use remote blast. Install a local copy.

[anikcropscience]

Thank you for the suggestion. Do you know of any NCBI database only for bacteria, viruses, and fungi? I have run Kraken and did not get what I was looking for since many of the reads were unclassified.

[dthorbur]

Which kraken database did you use? I agree that running a local version of BLAST should dramatically improve runtime, but I think I would alter approach a bit. If majority if your reads are not being classified as expected by a tool like Kraken, I would look closely into a handful of the unclassified reads. BLAST these, and look at all other databases available.

It sounds like you may have a contamination issue. If you used the correct one, the Kraken databases are very good at general classifications, so if samples are being unclassified, it sets off a few alarm bells for me.

[anikcropscience]

I used the Standard and Viral database from this source https://benlangmead.github.io/aws-indexes/k2

I have around 120K reads and 50% of those were classified and the other half remained unclassified.

[Joe]

NCBI databases aren't broken down in that way as such, you have to filter by taxonomic ID numbers.

Answer 1

[pbioinf]

I've had fairly pleasant experiences using BAT/CAT https://github.com/dutilh/CAT. It is Diamond based so should be faster than BLAST while using the same logic. You'll still need a decently powerful HPC for it.

[anikcropscience]

Thanks a lot for suggesting this.

Answer 2

[Mensur Dlakic]

For your specific problem, having larger memory, many CPUs and a faster disk are the only ways to speed up the process.

Another way of classifying contigs without BLASTing is to bin the contigs and then annotate using GTDB classification:

• https://gtdb.ecogenomic.org/
• https://github.com/Ecogenomics/GtdbTk

This typically takes hours or a day at most.

[anikcropscience]

Oh that is cool to know. Thank you. I will check it out.