Studying Pathogens Degrades BLAST-based Pathogen Identification

Using the BLAST algorithm to search the NCBI databases is the typical way one goes about identifying a DNA sequence, so it's been the typical way biosecurity systems decide if something is potentially a dangerous pathogen or toxin too. Problem is, that's not what BLAST and those databases were designed for, and we've observed that they aren't working as well for that purpose as they used to, as we report in our new preprint: "Studying Pathogens Degrades BLAST-based Pathogen Identification"

Specifically, we've found an inherent problem that is growing in seriousness due to a non-obvious emergent dynamic. Now that sequencing and bioengineering tools are getting much more accessible, lots of sequences are being studied by modifying them with "tool" sequences like purification tags, fluorescent proteins, stabilizing sequences, etc. Those sequences get (appropriately) classified based on what's being studied, and now you've got chimeric material that includes both the subject of study and the bioengineering tool. Then when you run BLAST on a sequence with that tool, you start finding that tools are classified as what they're used to study.

Example of BLAST classification failure: using a purification tag to study an Ebola protein means that now a fluorescent protein plus a purification tag gets mis-identified as Ebola.

This doesn't seem to be much of a problem for most uses of BLAST against NCBI, but it's poisonous for making biosecurity decisions, since it can cause benign sequences to be classified as dangerous or vice versa. Moreover, the effect gets stronger the more problematic a pathogen is (since more sequences are recorded) and the more useful a tool is (since more chimeric material is produced), meaning that the problem is most likely to occur in the most important.  For example, over the last two years, quite a lot of stuff has started coming back as COVID-19, since everybody in the world is studying COVID-19 with all of the tools that they can get their hands on.

This is a serious problem, and it's not likely to get better, since NCBI and BLAST aren't doing the wrong thing: they're just getting less suitable to use as a short-cut for doing something that they were never designed to do. 

So how do we fix it? Switch to tools that are actually designed for pathogen identification. We've got one (FAST-NA Scanner), and a whole bunch of other folks worked on the same problem in the FunGCAT program. The solutions are there, we just have to help folks switch to them.