Hello, and thank you for listening to the MicroBinfeed podcast. Here we will be
discussing topics in microbial bioinformatics. We hope that we can give you some
insights, tips, and tricks along the way. There's so much information we all
know from working in the field, but nobody writes it down. There is no manual,
and it's assumed you'll pick it up. We hope to fill in a few of these gaps. My
co-hosts are Dr. Nabil Ali Khan and Dr. Andrew Page. I am Dr. Lee Katz. Andrew
and Nabil work in the Quadram Institute in Norwich, UK, where they work on
microbes in food and the impact on human health. I work at Centers for Disease
Control and Prevention and am an adjunct member at the University of Georgia in
the U.S. Welcome to the Microbial Bioinformatics podcast. Lee and I are your co-
hosts for today, and we're again navigating the practicalities of bacterial
taxonomy. And we have some excellent guides joining us again. Our guests today
are Dr. Lee, who is a Strattide Chancellor's Fellow at the Strattide Institute
of Pharmacy and Biomedical Sciences in the University of Strattide. Dr. Connor
Meehan, who is an Assistant Professor in Molecular Microbiology at the
University of Bradford. So, welcome to you both. Thanks for joining us. Thank
you. Thank you so much for having us. It's been a while. So, we were discussing
last time a lot of the theoretical and almost philosophical issues around
bacterial taxonomy. I think this episode would be good to dip into some of the
more practical problems it is when we take taxonomy, as we previously described,
as something that is born out of antiquity almost. And then how do we integrate
that back in with genomics and what we're learning from genomics data, which
often is at odds with the classical taxonomic definitions. So, I think what one
of the first questions is, GTTB is incredibly popular. Donovan Parks is the one
who drives a lot of that, who was a PhD student when I was the postdoc working
with Ford. So, all of our taxonomy stuff comes from all these kind of
discussions with Ford Doodle and stuff. And then it's been an amazing amount of
work that goes behind that, and it just got updated. I think it's like now it's
defined as a database, not just as a concept, that this is something that will
now be continuing on. As you say, potentially a new paradigm. I think GTTB is
the first big, working, effective and widely used sign of where things are going
in terms of whole genome classification. It is, you know, it's a new paradigm,
genomes being the maximum amount of useful quantitative information that you can
get from all of the heritable material. And as we were discussing last time,
where we don't have phenotype or morphological information or any other
characteristics to go off, we only have genomes. We have to find a genome-based
way of classifying and identifying organisms. And GTTB is right at the forefront
of that. What's the secret sauce here? What has it done differently in its
definition? So, it may not be the first, but it's definitely the most widely
seen of where it's not trying to go back to that biochemical DNA-DNA
hybridization and try to find something that goes. Here are the genomes. Let me
build a tree based on these core proteins that are present in all of these
genomes. And then let me define some rank that go in there. So, we have
obviously the phylum and the class, et cetera, that are in there. And it uses
this normalization to try to say, in general, we see this split in the tree is
at this genus level that we've called a red, R-E-D. And all the splits that are
at this distance from the root, we're going to call them a genus. And all the
ones that are at this distance from the root, we're going to call them a family.
And so, it's a normalization way that is expandable, that you can add in extra
things, and it'll give you what that ranking is. Now, what it turns out, most of
the genera that were there fit within that. And most of the families that were
in there do fit within that. So, actually, whatever we were doing before, in
terms of defining it, has some basis in genomes and some basis in phylogeny. So,
this is really a phylogenetic-based taxonomy, instead of trying to always rate
it back to a phenotypic-based taxonomy. I think it's moving away from
nomenclature as well in a really useful way. So, one of the things that comes up
in presentations I've seen is that of all the taxa in GTDB, about 80% don't have
valid names. And GTDB is kind of notorious for having reclassified some
organisms and produced names that didn't follow the nomenclature and the
prokaryotic code, causing a few arguments. There's more of those. Thankfully,
when the mycobacteria was split in five, the GTDB said it should all be one, and
that was front and center on my paper. I'm glad they could resolve it in that
case. Yeah, that was good. But there are some other ones I can't remember
exactly, but definitely where they split it, and not everybody was happy about
that split. It comes back to the usefulness versus the biological, quote-
unquote, truth. And the use of Latin as well, going back to the historical
overview of it, the nomenclature. Some of the Latin derivations of the names
were different to those that would have been used in terms of prokaryotic code.
And I think GTDB is a useful and welcome step away from having nomenclature
govern what's acceptable in terms of taxonomy. I think we're only going to move
further in that direction. Yes. But how do you feel about the sticking with
phylum order class genus? Because I would have thought that with the huge amount
of genomic information we have, we could look for where the natural breaks and
discontinuities are. We don't need to be constrained by that old set of
taxonomies. So I think that's where it's going to go. What the GTDB allows is
for a lot of meta-analysis for people who want to ask those kind of questions in
the phylogenies. I'm thinking about Tom Williams and doing those kind of, what
does it mean to be a phylum? And now you have all the data that's out there, and
then you can really define them. And then it's a, where is taxonomy useful? And
where is this something where evolution took a turn and that turn was not
reversible? And that has defined this group in some way. So I don't know what's
going to happen with that, but I'm not clever enough to do those papers, but I
look forward to reading them and then going, oh, that was really simple. So in
terms of those higher ranks, what is interesting that I think a lot of people
don't know is that because it's my first lecture in introductory microbiology is
taxonomy. And I say, these are the ranks. So everyone takes it as gospel. But
the definitions of those ranks is not a definition. The official definition of
genus is it is the rank above species and the rank below family, and that's it.
There is no other way that it is defined. But if we think often when people say
a certain genus name, those that work in it go, oh, clearly it's because it has
this. Its cell wall is this way, or it has this set of processes or infects
these certain things or lives in these certain environments. But there is some
kind of potential rank and grouping that's there, but whether it's a sampling
bias and we haven't just sampled everything around us, which we see much more in
virology, or whether there truly is a biological separation of these ranks, I
think is still to be seen. That's a great point. All of those, all
classifications, well, many of them are based on a real undersampling of what we
now know the biological diversity is. And to have those set in stone and define
where subsequent classifications as we sequence these organisms go, I think it's
probably something we could revise usefully. And having, I keep raising the
point, but having a qualitative approach to this and having those kinds of
categories for organizing how we look at genes, distinct from how we refer to
them in terms of cases like E. coli versus Shigella, where we have the societal
need for clear, distinct names that mean something for real purposes, whether
it's pathogenicity or whether it's utility in an industrial context. I think
it's fine to have those parallel schemes, but we need to not be beholden to
nomenclature. Am I saying that too much? Every time you mention it, take a
drink. For those of you listening at home. It's illiterate, I am not getting
drunk. So in theory, since we're going to talk more technical on this one and
not go back too much into the philosophy of it, what the GTDB and a lot of other
people are doing is trying to give a technical way to at least define what we
see and to be disputed. So a lot of work has been done on defining the species.
And the way that it really is done probably by most people now is you have all
of the species that are out there. You have your new genome that you think is
the type for what you're working on. You calculate the ANI, the average
nucleotide identity between them. There is not just one way to calculate that.
There is fast ANI and other ones that will give you slightly different, but
almost the same answers. And then if it's 95% or above, it's the same species.
And if it's lower than that, it's not. We tend to give it a little bit of a 94
to 96. And in theory, a subspecies is 98. Not always kept that way, but that is
one way to do it. There have been ANI barriers that have then been created for
the other ranks. They don't hold anywhere near as well. There's a lot of
overlap. Even part of that is because the genera are all defined differently.
And part of that is because it's just a nucleotide identity. When you have back
transitions, it doesn't hold up the further back in evolutionary history you go
because of your back mutation issues that we spoke about on the genetics podcast
that you can go back and listen to. Yeah. Even in identifying homologous
regions, which is most of the time what you're doing with ANI, there's kind of a
limit of 70 to 80% identity for any pair of aligned homologous regions. Just
algorithmically, you can't detect distinction from random identities lower than
that. So one of the things you can use to get around that is coverage. So what
you're looking for, very similar homologous regions, but you're looking for how
much of each genome is covered. And one of the things that seems to arise from
that is that you can get quite high ANI scores, so percent identity, but you get
quite low coverage values, 30% or below, which seems to be a reasonable way to
distinguish between members of the same genus and members of the same species.
So you're looking for high ANI identity, but you're looking at relatively low
coverage.  use 50% because if you're sharing less than half of your heritable
material, maybe you're sharing more than half with something else that isn't in
this comparison. So I tend to be fairly conservative about that as a rule of
thumb. But there's a really good paper out, I'm forgetting the author, I should
probably look that up, where they looked at a number of different groups and
looked at ANI identity and coverage, and depending on the groups, somewhere
between 30 and 50% coverage seem to be a good delineator of same genus but not
the same species. So a lot of work has been done on the species level and now
work is being started to done on the genus level. So that's been something I've
been working more on because in mycobacteria, as we spoke in the last one, you
got split into five different genera. And most people, this had been done by
some authors and other groups, but then they touched a group that's special to a
lot of people, which is the mycobacteria, because of the amount of clinically
important strains that are in there. And it's kind of seen as being one of the
most clonally, it doesn't have a combination, kind of everyone's like, how is
this not a genus? This is one of the definitions of genus that we use, because
it's defined by this mycolic acid cell wall, it's very clearly separate from all
the things that are around it. And it splits up when you use something like
average amino acid identity, the amino acids equivalent of ANI, AAI. But the
difficulty is that there's not a lot of other ways to define the genus, but now
people are really working on this. So Roman Barkow had a great one that we used
in our paper for showing that it was one genus, where you have the average
nucleotide identity and the alignment fraction between things that are
bidirectional best hits. So you're essentially using BLAST to say these two
things are equivalent. And then what is that alignment fraction between the two
of them, then you can kind of create this inflection point on your XY graph of
ANI alignment fraction. And that tends to say, you do that to the type species,
for the genus. And then you say, if it's closer to that one and further from the
type species for a different genus, it's probably all one genus. So we've got
very good technical things coming in for species that's moving up towards genus,
but nothing is really above apart from the GTTB way to go family and above.
Hopefully people will start looking at that more now that you have the genomes
and we can do some interesting things around that. I think that speaks to a
general trend as well, which is that when we look at ANI and other measures, and
we have these threshold cutoffs, 94 to 96% identity, I think all of these single
measures, not adequate really to describe the distinctions between species and
other groups. And moving to a combination of identity and alignment fractional
coverage is one way to move towards two measures in ANI. But having the mycolic
cell wall, taking multiple gene trees and looking to see where the consensus is
for sort of last common ancestor with it for multiple genes, not looking at a
multi-gene tree where you concatenated sequences, but looking at them
individually. All of these lines have ever useful to help support or refute the
assertion of a species boundary on the basis of ANI or some similar single
measure. And I think that for robust species classifications, we're going to
have to take those approaches. Obviously, we can't do that for the complete set
of sequence genomes. So we're still going to have a bit of conflict, even if we
do go forward to a completely qualitative and computational way of classifying
these organisms. I want to change tack a bit now that we've been talking about
the complexities of going species and above, let's talk about species and below.
Simple things like how do we then define a strain? What is a strain versus an
isolate versus a culture versus a variant versus a lineage? Are these something
that we can define and we can talk about? I get in trouble for using strain and
isolate interchangeably in my papers by the other authors on those papers. So
definitely it's not an easy topic. What I would say is taxonomically,
officially, if we go back to official, there is no definition below species
apart from subspecies. And even then that's quite dicey. But the term variant or
lineage has no official nomenclature standing. Anybody can create a variant.
Anybody can create a lineage. So it's more about trying to decide on what it is
that you're trying to convey with that. Isolates tends to mean, we took this and
we sequenced it directly from the sample. So the sample is from the soil, let's
say, and then the isolate is exactly what you did really without culturing it in
between. And then the strain is what came from that isolate potentially when it
was cultured. That's one way of thinking about it. And I'm sure there's many
people who are listening will be like, that's not how I would define it. And
that's also true, the difficulty there is. Please add us on Twitter. Just go for
it. Let's start to play more on it. To be internally consistent, I try to use it
that way. Clinically, I was saying this is directly what came from the patient.
And then this strain is what we've been using that maybe is a laboratory strain.
But people also define strains as like wild type strain, as these are
susceptible and related, or they share some kind of virulence pattern and are
related. So the word strain is often the difficulty. I think people kind of know
what they mean when they say isolate, and they kind of know what they mean when
they say lineage, and then strain is used in both of these places. I'd say in
conversations, I've heard it used not necessarily interchangeably, but
differently by different people, as well as there being no real taxonomic
definition, there's no formal definition. I don't think everybody uses it
entirely consistently, but I think I've got the same kind of general feeling,
the same sort of head count and gonna, that the isolate is that the physical
thing that you took as a sample to some extent, and that you sequenced, and the
strain is the lineage that comes from that, everything that shares it as a
common ancestor. So if you've got a strain in the fridge or the freezer, then
what you sequence 20 years later is possibly not what was originally in there.
So you're sequencing that strain, but you're not necessarily sequencing the
isolate, which was the physical sample that you took from the field or the
planet 20 years ago. And I tend to try to avoid the word variant as much as I
can, just because we use variant to mean mutation at the same time. And there's
already enough confusion between a variant and a mutation in terms of that
technicality, but then trying to use variant for the whole thing. It's
interesting having lived to COVID and then being involved with the COVID
analysis and seeing the evolution of the term variant, because it's like, yeah,
anything with a single step difference, that's a variant, right? But some
variants are more variant than others, like, you know, they're the same, but
they're not the same, because then you're interested, well, then which one
should we be concerned about? Variants of concern, variants of interest. And now
we've got this tiered level of how panicked should we be about every given
variant. So I would tend to avoid using the word variant now in the same way I
would have tend to avoid using the word pandemic, because I'm sick of hearing
it. Yeah, so I mean, viral taxonomy is a whole separate thing that we could talk
about for a long time. And I'm definitely not an expert. There are people at the
Quadrant, especially, who are much better at that. It used to be this term
subtype that was used. If people work in HIV, there are a lot of subtypes of
HIV. And then that term is kind of no longer being used. Because then it was
circulating subtypes and circulating recombinant subtypes, and all of these
kinds of things. And virology, they've moved in towards variant. And I think now
with SARS, people are moving more towards the term lineage. And in bacteria,
there's also a move towards the term lineage. How I often like to think of a
lineage is, if we look back at this in a million or 2 million years, we'd say
that's where the species came. There was enough of a separation going on that I
can trace that back to what was a lineage inside one species at some time. There
should be some kind of difference that we care about, that is a monophyletic
group being on the phylogenetic tree, but also has something that differentiates
it from everything else in terms of its virulence, in terms of its ecotype or
infecting, whether it's host range or environments or something like that. But
it should have something that you can define outside of a set of SNPs. And
virology taxonomy is really leading the front with that kind of thing, because
they evolve so quickly that they get lineages. They get multiple lineages in a
year, as we've seen, whereas for bacteria, you're waiting thousands of years for
these kind of things to really come apart enough. And then obviously, if you're
going from salmonella, you've got a virus, and what does that mean in relation
to a lineage in relation to phylogenetics? And it can get more complex there for
sure. I'm certainly no expert on viral taxonomy, but watching what's been
happening with COVID has been fascinating, and it's really affected the way I
think about how we should approach bacterial classification and nomenclature.
It's, as you say, really compressing what we're seeing bacteria into a very
short timescale, and there are direct analogies to be drawn in terms of lineages
and what really counts as being distinct, which has changed what I think. And
we've seen that in the pandemic, as you say, we started out with variants of
concern, but then using the word variant is so close to the variation that
people then just got freaked out when any variation occurred. And then you're
trying to get into an evolutionary background of like, oh, but is it due to
selection pressure or genetic drift? And then media is like, I don't care about
any of that. Tell me whether I should be like in a lockdown again. A lineage
should be generally because there's pressure and in terms of selection that's
occurring there. And we see that now with the alpha and the beta and the delta.
And we see that with bacterial lineages where there is some separation and
virulence normally within a species, but taxonomically it's not defined yet
because it's really dependent on what you care about inside of that species and
what should be reported to the clinician or the environmental officer or
something that you need to care about this. Or like you said to us before with
Listeria, the opposite direction where you actually want to be grouping a lot
more things together because we care if it's any of these things. We don't want
to go down to the minute level. We actually want to care about the thing that
causes the disease no matter where it comes from. Could you just contextualize
the Listeria story you were telling us earlier? That was probably before we
started recording, but this discussion in terms of, in the context of public
health, which is all things for me, but I mean, when you look at Listeria, I
think it has an ANI of like,  like 92% across all lineages. It has four
different lineages, one through four. Lineages one and two are more associated
with disease with people. Three and four are associated with animals. And even
though three and four cause way less disease in people, it's still all grouped
together. And that's, I would say that's primarily because of regulatory
efforts. If you find any one listeria in food, then it's been subject to recall
and public health actions, probably the same over there in UK. And it just, it
makes me think, okay, we have an academic versus a pragmatic definition. We have
an academic definition of around 94 to 96%. I like how you have a little wiggle
room, as you said before, versus 92% across the species. But pragmatically, it's
for the greater good to make sure that we just don't have listeria in our food.
And a lot of taxonomy comes down to that pragmatic. And now I think people are
trying to come at it from a biological point of view. But where I think a lot of
taxonomists often sit now is saying, okay, I know you want to destroy the system
and put in a new system. But for a lot of people, their entire living depends on
knowing the system very quickly, being able to regulate it and all that. And I
know I come back to mycobacteria a lot, but one of the reasons that we did the
genus paper was because we were contacted. One, we were the mycobacteria
collection. Did we have to now change our name to be the mycobacteria and the
mycobacterioides? But it also was the diagnostic tools that are used by Public
Health England, by the CDC. They say that they detect mycobacteria. And if they
no longer detect mycobacteria, but actually detect 5-genera, did they have to
spend, did these companies have to spend thousands in regulatory ones to have
the inserts of their products changed? So it came down to, we did that for a
biological reason, because I was like, this is not right, but really it's driven
by a pragmatic reason of do we have to change regulations to say that this is no
longer this? And that has huge, wide ranging implications. Yeah, there are
essentially three branches of taxonomy. There's classification, there's
nomenclature, and then there's identification, which is where legislation meets
taxonomy. And not just in terms of the validation, demonstration of test
accuracy that you bring up, but also trying to get new legislation through if
organism names change. Changing policy is a slow process. Changing legislation
is slower on most occasions. If you needed to put into law that you suddenly
needed to quarantine all imports of potatoes that contained these 5-genera
instead of the single genus, it would take quite a long time. And during that
time, you're at risk of introducing a number of phytosanitary threats. So the
very practical and applied nature of taxonomy really comes out with that
legislation. And also when you talk about public health as well, it reminded me
of pathovals, which we use to discriminate below species level in pathogens,
which reflects the susceptibility or otherwise of plants on exposure to the
pathogen. And that might be mediated by the presence or absence of a single
effect that gets detected by the host plant. And there really is no great
genomic distinction. Certainly you call these organisms the same species, but
they might differ quite drastically in terms of where they cause disease.
Historically, they've been classified as pathovals and considered to be
different. And they represent different threats. And we need a scheme that
accommodates that and allows us to name them clearly so that we can structure
legislation and the societal effects of those organisms. So I don't want to
belabor the point too much, but the example I always give about taxonomy when
it's like, nobody cares until they care. And the example I give normally is the
polar bear and the brown bear to everybody are very, very different bears. And
this has massive wide ranging because the polar bear is protected and the brown
bear is not protected. They are the same species. And recently it was shown that
they are the same species. But if you do that, and this was nearly caught on by
governments, they were like, oh, so then we don't have to protect the polar bear
and all the money that we put into it because it's no longer a species to be
protected because it's now part of the brown bear, so it's fine. So it's nobody
cares until they care very much so that the polar bear was just going to
potentially lose all of its protection because it was decided that taxonomically
then they're different. So it's an area where the technicality, to come back to
what we're talking about here, the technicality is really coming along of how we
define it, but we're trying to bring along with that an insurance that it
doesn't override the practicality. Technicality needs to support the
practicality in both directions. And if we come back to COVID, obviously the
lineages are very different in terms of epidemiology, but in terms of public
health, one could argue that they're the same. We should be doing the same
things to protect ourselves from alpha that we do from delta, which is mask
wearing, et cetera, et cetera. But people think of them now as completely
separate things. People think of COVID in 2019 and COVID now as not being the
same thing at all. And that's really about the taxonomy because we've said
delta. We don't call it SARS COVID, we call it COVID and we call it delta. It's
still the same thing and we should be applying it in the same way. So
technically how we define these lineages, I think is going to become a big area
of research going forward because now we really see in public health how those
definitions can have very wide ranging impacts. I want to change tack if that's
okay, since we've talked about different technicalities. I want to talk about
some of the tools that people can use to apply taxonomic principles to their
genomics for bacterial genomes. We've touched on ANI and we've touched on AAI,
so average nucleotide identity, average amino acid identity. What else is out
there? Oh, there's more than one version of ANI and it does matter which you use
because they can give different results. And if you're applying a threshold,
they can put you on side or over that threshold. There's digital DDH as well,
which is actually worth discussing in the context of ANI because there are some
limitations to both of those approaches which they have in common. So digital
DDH was intended to essentially model actual in vitro DNA-DNA hybridization. So
if you go to the GGSC sites, put your genome in, it will go through a model
which was fitted to outputs from DNA-DNA hybridization. One of the problems with
doing that, in my view, is that when you compare outputs from in vitro DNA-DNA
hybridization and you look at the actual level of genome identity in the
pairwise comparison, there isn't a strong mapping. You can fit a straight line
through it, but it's got quite a lot of variation, particularly at the lower
end. And if you're at 70% DDH from the in vitro measurement, you could still be
up at 95% genome identity when you do the pairwise comparisons. So fitting a
model to try and predict, to me, always seems like trying to hit a very broad
target and people tend to take single numbers that they get from a computational
tool and believe that that number is accurate. Now, a lot of the time, it does
give you values which are in line with what you'd get from other methods like
ANI, but I'm always a little cautious about using it because there isn't this
strong relationship. There's a strong mapping between DDH and ANI. ANIB,
versions of ANI that use BLAST, try to emulate DNA-DNA hybridization in a
similar way, but what they do is they fragment the genomes that you're trying to
compare. So they chop them up into arbitrary sections. Usually it's, if you
follow the original paper, it's fragments of about a thousand nucleotides and
they look for homologous regions between the query and the subject genome, but
they're looking for a minimum match along the length of the, which means that
you could be losing some information about homologous regions when you make your
comparison. This results in a difference between the ANIB approach using BLAST
and the ANIM approach, which uses MoMA and tries to identify the maximal
homologous regions, particularly when you get down to low percentage identities.
Above 95%, they're usually very close, so not really that important, but the
further you get away and you're trying to look at distinctions that are beyond
species level, I mean, it does start to matter more. And then FAST-ANI is an
attempt to approximate average nucleotide identity calculations, which are
obtained by pairwise alignment using K-mer approaches and MinHash. They do
really good approximations incredibly quickly, but it's often difficult to get
coverage information back and you can't see exactly which bits are aligning. And
if you've got a small set of genomes, sometimes it's useful to just have the
alignments there so you can go and look at them and see what's contributing to
this similarity and if that's phenotypically or functionally important for you
and your studies. As I was going to say, there's a lot to unpack there,
especially if you're a student starting out and you're trying to get on top of
these tools. I mean, those tools are definitely, when you want to do taxonomy
work of defining things, then there is a whole suite that's in there and the
nuances are in there, and obviously phylogenetics plays a large part in there as
well. But I would say most people, when they think of taxonomy these days, it's
I have my metagenomic sample and I want to get the taxonomy of what's in there.
So practical taxonomy these days is centrifuge, mash, like trying to say, what
is it and what do I care? You know, MLST and all of these things. I want to put
a label on what my sample is and then I want to know what that label differs
between these two different things. So practical taxonomy is taking all of these
genomes that somebody else has put labels to and then finding out which of those
labels should be applied to your sample. And I guess what myself and Leighton
work on is trying to make sure that the labels that are coming at the top are
the right labels that are going to come down to you and making sure that I'll
make some kind of sense so that you can say that, well, I see this strain
lineage species genus here and I don't see it here and that means something. But
really when most people are doing taxonomy, they're running QIIME and certain
such tools like that. And I think in the last couple of minutes, just wanted to
ask about how both of your views on moving away, I suppose, from traditional
taxonomy, but more to just a simple.  nomenclature problem of how do you define
clusters and define groups that you can describe to other people and do it in a
systematic way. This comes back to kind of what GTDB is doing for, doing for all
microbes, but then what about within a species? I'm interested particularly in
the work with genome archives, which is a whole genome database for classifying
microbes. And it reminds me a lot of PHE's SNP address system, where in both
cases you have this sort of threshold and then based on whether something was
similar at that threshold, it would have the same identifier. And so something
that shared a lot of different numbers would be closer, more closely related.
Something that shared a lot of identifiers would be less closely related. Let's
say that the right way around. And it reminds me of the, and then I'm going to
throw in CGMST because I love CGMST. It has a similar sort of process that you
can make distinct classifications based on that. How do you both, how do you,
two of you feel about these tools and what's your experience? When I describe
the work that I do, I often say, oh, I work on epidemiology and I work in
taxonomy and I work on these different things. The reality is that I liked
cluster things. And then when it's transmissions that I'm clustering, they call
it epidemiology. And when it's everything above that, then people now say I'm a
taxonomist, but it's essentially the same thing. When I'm doing my epidemiology,
I'm trying to look for clusters that are meaningful and separate from other
things that has an actionable thing at the end. Is this transmission cluster in
this place? And it is different from this transmission cluster in this other
place. And therefore we can say something about that. But epidemiology is just a
public health version of taxonomy. There I said, but if you enjoy working on
defining different clusters, then you're essentially doing some kind of taxonomy
with data set that you're working on. And then every, and then it's just about
what's useful. So the SNP address works really well because everybody knows what
they mean when they say that. And when I say transmission cluster, everyone kind
of knows what I mean by I say that I say how it's defined, but then once I've
defined it and everyone agrees that the way I've defined it as a valid way to
define it, then we can just move on with our lives. So it's just about creating
a nomenclature that's going to work. And it just happens that the binomial
system is the one that is taught at the university level. So we're all ingrained
into us to have that as researchers. But I think the rise of MLST and especially
CGMLST is defining the taxonomy of epidemiology, which is what we're kind of
doing a little bit more now. What does that mean? And then that grows somewhere
into a lineage. And that's where these two things collide, where
epidemiologically we've defined these lineages. Honestly, they're transmission
clusters that have grown around the world. And now we've decided to put what we
think of as a classic taxonomic label onto that. And it's all just a spectrum.
What we're trying to do with Genome Archive is really what Colin was describing
in terms of having that framework for where you can identify a location for your
genome or set of sequences that you've obtained from your sample. It's one of
many possible and valid and useful frameworks. And it's grown out of Lynn's life
identification numbers, which were proposed by Boris Vinatsa and Lynwood Heath,
who are both at Virginia Tech. And the way I think of it is that it's a
hierarchic subdivision of genome space. You've got this abstract space, and
every individual genome is a point in that space and is in some way addressable.
And what we're trying to do is define volumes of space that enclose these
genomes. So you can say which genomes are in which space. So we have a label for
those spaces. And you can kind of think of it as building a tree on the basis of
a distance measure. So as a first pass, you could use ANI and calculate ANI
distances and produce a distance tree. If you lie it on its side, so you've got
your root on the left and you've got your leaves on the right, you can then put
an x-axis in and take vertical lines which partition the tree at various points
between the root and the leaf. And we can label those A, B, C, D, and so on and
so on. And each of those partitions, you can then count from the top down to the
bottom the number of individual groups. You can label those 0, 1, 2, 3, 4, and
so on. But keeping in mind what the last classification was just off to the
left. It's difficult to show without a whiteboard. I really want a whiteboard to
show this. But essentially what it gives is a unique index at a number of levels
between the root of your tree and the leaves of the tree where you can define
essentially a place in the tree and label that consistently. That maps onto this
sort of abstract space where you've got each genome as a point where these
individual labels don't describe locations on the tree but they now describe
like voxels, if you remember, sort of like graphics and computer games. So every
voxel has its own label and within that each of the subgroups voxels has its own
label and so on. And it drills down so you get down to an individual or at least
indistinguishable genomes. And this makes it very much like a map grid
reference, like an OS map. So if you had a label like, say, that contains 15,000
genomes, that's like a big sort of grid that covers a country or a city. And if
you drill down further over to the right, and you're getting closer and closer
to individual streets and you're gradually grouping things which are closely
similar and excluding things which are not quite as similar. In terms of the
distance measure that you started out with, one of the issues that we have with
that technically is going from the root of a bacterial tree where we can't reach
that with ANI, as we talked about, getting to the leaves which we can't reach
with ANI because you end up not being able to distinguish between individuals
that are distinguished by single snips or you've got really closely related
lineages in an outbreak. So what we're trying to do is take multiple distance
measures and match them up so that these hierarchies are consistent. And we can
use things like split k-mers down at the leaf end for outbreaks and we can use
AAI at the root end for sort of phylum and order level distinctions. And one of
the ways I like to think about it is that it's sort of a Rosetta Stone and a
neutral reference. It's like when you're looking at a map of, say, a disputed
border, like between England and Wales, that's shifted over history. But every
point on that border has existed throughout the history where it's moved more
towards Wales, more towards England. So you could take that point and a period
of time you could say whether or not this is Wales or this is England. So it's
got, for that individual label, a number of annotation states which are
associated with it. So we can hang multiple different nomenclatures from the
same framework reference. So we can label each of these divisions with
phenotypic or host specificity or some other useful biological information that
we know about. And if we have a volume, one of these labels, which contains
genomes which all share the same characteristic, a new genome which you bring
along and then finds itself in that volume, you might reasonably expect to share
that characteristic. Or if a majority of the genomes in that volume share the
character, you might expect them also to share that characteristic as a
probability. And just as with grid references where, say, for Manchester, you
wouldn't expect a square to exactly match Manchester. You could match Manchester
to an almost arbitrary degree by specifying collections of grid references of
large or small size and then grouping them together and saying my definition of
Manchester is this. And it will describe the outline of Manchester almost
exactly. And I think we can do the same with voxels in this sort of genome
space. So we'd say this species consists of this label and also these other
labels. And we would have a nice consistent frame of reference for that
definition of that species at that point in time. And that could act as, say, a
Rosetta Stone for translating between nomenclatures like uncode and set code and
traditional binomial nomenclature and the historical references to nomenclature
through literature. The other advantage, well, the Genome Archive project itself
is trying to build this. And we want to build a usable resource which is
available to all and also privacy focused. So for considerations such as
indigenous rights, if there's any IP that you want to protect, we want to be
able to allow people to not upload the exact genome sequence, so they protect
the privacy and their rights to that organism and that sequence. So we want to
have the interface such that you can process your genome locally in the browser
and then the kera profile of that sequence is used to locate your organism
within that space, which is something that you can really only do when you take
this computational approach. It's a big and challenging project and it's really
exciting. I apologize if I don't sound quite excited enough, but I'm optimistic
that it's a step in the right direction for how we can consider organizing
genome data and relating it to all of these tensions and conflicts in
nomenclature and taxonomy and the different ways of viewing these organisms. The
same way you can look at a tree and you can focus on the trunk or the leaves or
a particular plane through the tree. There are so many ways of looking at this.
This is only one of them and we think it's moving in the right direction. And I
should say that this is, just give everybody credit, it's Boris Venatsa is
leading it at Virginia Tech, along with Leonard Heath at Virginia Tech, it's
Davis, my part's over here at Strathmore. And what we will do is, there are
excellent posters for this up on Figshare, so we'll put a link to that in the
show notes if people want to go and have a look and dig into this a bit deeper.
I think we're almost out of time, so I have one final question for everybody
when it comes to taxonomy. Are you a lumper or a splitter? I'm a splumper.
That's a binary that doesn't actually really exist. Sorry. Sometimes lump,
sometimes splitter, it depends. I'm quite pragmatic about these things. I might
sound like I'm ideologically driven too. When it comes to taxonomy, it's more
about, are you a...  biological purist, and you want only the biology to drive
it, or are you pragmatic with it? And I think a lot of taxonomists are quite
pragmatic of saying we should lump when it's important and we should split when
it's important. Oh, fair enough. Okay. Well, that's all the time we have for
today. I want to thank our guests, Dr. Leighton Pritchard and Dr. Connor Meehan.
We've been talking about vaxillotaxonomy and some of the practical genomics
tools available. And we'll see you next time on the MicroBinti podcast. Thank
you so much for listening to us at home. If you like this podcast, please
subscribe and rate us on iTunes, Spotify, SoundCloud, or the platform of your
choice. Follow us on Twitter at MicroBinti. And if you don't like this podcast,
please don't do anything. This podcast was recorded by the Microbial
Bioinformatics Group. The opinions expressed here are our own and do not
necessarily reflect the views of CDC or the Quadram Institute.