19 - Misunderstanding of BLAST max_target_seqs setting and implications for programs with similar behaviour

A recent paper by Shah et al., 2018 explains the default behaviour in BLAST using the --max-target-seq setting. Many people use this setting to restrict results to show only the 'top BLAST hit' when make taxonomic assignments. The documentation explains that the first N hits found in the database are returned. Most users think this means that the first N 'best' hits are returned. In fact, the first N hits with an e-value better than the cutoff are returned and are not guaranteed to be the 'best' hits. Link to paper:

https://doi.org/10.1093/bioinformatics/bty833

This parallels an issue with QIIME that uses UCLUST for open & closed reference-based taxonomic assignments. UCLUST makes assignments to a reference set. These assignments can change based on the order of sequences in the reference set because assignments are made to the first reference sequence that passes threshold cutoffs, not necessarily to the best reference sequence. Link to Sebastian Schmidt's thread:

https://twitter.com/TSBSchm/status/1049000995395702784

What can we do about this?

Stop using the top BLAST hit method for taxonomic assignments. There are numerous alternative methods that are faster and provide statistical measures of confidence for all ranks of a taxonomic assignment, e.x. RDP classifier.

Consider not clustering your metabarcoding data into OTUs at all. One alternative approach is to simply 'denoise' your data. See post on 'OTUs vs ESVs'. If you really need OTUS, then try an alternative clustering strategy that avoids arbitrary clustering thresholds and input order dependency such as SWARM (Mahe et al., 2014).

Here is the reply to the paper by Shah et al., 2018.

References

Madden, T.L., Busby, B., Ye, J. 2018. Reply to the paper: Misunderstood parameters of NCBI BLAST impacts the correctness of bioinformatics workflows. Bioinformatics, Accepted.

Mahe, F., Rognes, T., Quince, C., de Vargas, C., Dunthorn, M. 2014. SWARM: Robust and fast clustering method for amplicon-based studies. PeerJ, 2: 3593.

Shah, N., Nute, M.G., Warnow, T., Pop, M. 2018. Misunderstood parameter of NCBI BLAST impacts the correctness of bioinformatics workflows. Bioinformatics, In press.