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. Both
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. So I'm Andrew. I'm at the Microbial Bioinformatics Hackathon
here in Bath, and I'm joined here by a few guests. Torsten. Hi, I'm Torsten. I'm
from the Doherty Institute in Melbourne, Australia. Hi, I'm Finlay McGuire. I'm
at Dalhousie University in Canada, and the Shared Hospital Lab in Toronto. And
I'm Christy Horan, and I work at MDU in Victoria, Australia. Okay, so just
before this, we were talking about, you know, some interesting things you've
come across, and you had one involving PDFs. Could you tell us? When I first
started as a baby bioinformatician at MDU, I started having to validate AMR
versus phenotype. I came across a fantastic, very large data set that I was
going to use, but needed to get the phenotypic metadata so that I could compare
to the genotypic results that we were seeing. And the only available information
I could find was a PDF. So not like a database or anything, or access database?
No, no. Not even a spreadsheet. It was a PDF with very useful information in it.
And so I had to learn how to parse a PDF of 10,000 samples or so into a
spreadsheet, and then put that into some sort of useful format to compare to all
of the genotypic results we were seeing. I must admit, it did occur to me that
maybe I might have taken up a job I might not actually want in the future, after
having done that earlier in my career. Well, at least it wasn't handwritten, you
know, that would be many times worse, like. No, I think that's the only way it
could have been worse, is if it was actually handwritten. Yeah, absolutely. And
what about your phen? I mean, I think they have now added a different approach
now, but I think until relatively recently, one of the only ways to get all the
cut-offs for antibiotic susceptibility tests for MICs to categorical cut from
one of the large providers was as a large set of PDF tables. They were all
inconsistently formatted as well, slightly differently. It was almost impossible
to automatically rip out, it was quicker just to take down the data. So, was it
ever manually made, like in Word or something like that? No, it was definitely
extracted from Excel, it's just they were distributed as PDFs. That's nice. I do
remember submitting stuff to a journal, you know, like session numbers for
thousands of samples, you know, and then you click, you know, build PDF and it's
like a 300-page submission. I was like, maybe you shouldn't do that, you know?
My main PDF story, though, is in that kind of disconnect, as bioinformaticians,
you know, we generate these datasets, generate these reports, and often there's
some sort of spreadsheet report, but there's usually a PDF report with nice
figures in it, and I assumed that the spreadsheet was being used and
automatically ingested into, you know, an electronic medical record or what have
you, until a feature request came in of, the final summary table in the PDF,
could I make the font size bigger? I was like, what? Sure, I can make the font
size bigger, but why? Oh, when we print it out and type it into the system, the
font size is a bit small so people can skip rows by accident when they're using
the ruler to mark them. So we made the font size bigger and it reduced our error
rate in data creation. Yeah, that reminds me, actually, one time I was asked to,
we had this phylogenetic tree which had 2,000 strains in it, and I was asked,
could you print that out? And they were like, this is too big to print out, but
anyway, we were forced to by, you know, an academic, and then it turned into,
you know, this thing that was, you know, a metre wide by, like, a few metres
long, and we took a photo of it, actually, because it was so impressive, and
that was just to get it slightly big enough so you'd get, like, size 8 font, you
know, to be able to read these strains. Yeah, we were not asked to do that
again. But you had something similar as well. Oh yeah, I mean, first, there's a
picture from, like, first week of my PhD with some phylogenies long, and so I
was looking at the evolution of folate by gene fusions in the eukaryotes, and
eukaryote tree of life polarisation stuff, and yeah, they were such large
phylogenies that they were basically stuck into the ceiling, and they hung to
the floor, and my PhD supervisor just would work his way down them and go, that
shouldn't be there, that shouldn't be there, and mark it up, and then we had to
go back, check the alignment, we did a few rounds of that. Yeah, anyway, thank
God we have better tools now, you know, you can zoom in on trees and all that
kind of jazz, like, Jesus. Click on ancestors. Yeah, yeah, interactive. Run out
of RAM. But actually, you know, just looking at the overall structure of a tree,
it does help a lot, because often you can see areas, you know, screwed off,
like, usually things are crazy, and you think, yeah, that's not right. Yeah.
Torsten, what about you? Yeah, in a similar vein to what you've described here,
I'm thinking back to about 10 or years ago, where we started working with a
large external organisation, and they had a computer scientist guy, he didn't
really have any experience in biology, but he was getting involved in a material
project, and he loved k-mers, everything to him could be done with k-mers, you
know, he just discovered them and thought they could do everything. They can't.
And so we were working on Salmonella, I think, and he decided to do a dot plot
of all Salmonella in GenBank versus all Salmonella, using k-mers as like an
identity measure. Yeah. So a very diagonal dot plot, right, not a lot of stuff
outside, but we had a person to person meeting at the end of the week, and he
pulled out his briefcase and took out this piece of paper with lots of sticky
tape on it, and then he proceeded to unfold it all over the conference desk. It
was A4 sheets, about probably 30 by 20, printed out, and he sticky taped them
all together. And he laid out this dot plot in the middle of the desk and said,
oh, here's all the genomes compared. And literally, it was just a diagonal with
a few, you know, there were some repeat elements, there might have been some
little bits off the diagonal, and this was his grand presentation to us all. And
we all just looked at him dumbfounded and didn't know what to say. I remember a
PhD student, and he was, he had 700 genomes in his data set, and he went into
Artemis, into like this genome viewer, and looked at all 700 manually to look
for a particular region, and was like, are you sure you want to do that? He was
a human blast. Eyeballed 700, which, okay, you eyeball a few, you know, to make
sure things are going right, but that's when you use a program to automate
stuff. Yeah. Yeah. So then during COVID, we found things that were really
annoying, not really annoying, but inconsistent with date formats. You get every
single possible date format in the entire world came into our lab, because every
system, I put it slightly different, you know, like the year is two numbers, the
year at the beginning, the year at the end, and no one ever knew what was
consistent or not, and you had to kind of go and manually reformat it in the
spreadsheets. The U.S. date system versus the rest of the world is one of our
pet peeves. Yeah. And if you're lucky, the date in the column is actually the
date it's meant to be, let alone the format. That's only if you're lucky. Yes.
Data collection, data sequencing, date of birth of the patient. Which one is it?
Yes. COVID wasn't around, you know, in 1921. Yeah. No, there's a lot of those.
And, you know, with the Kogu K project, I know that so much of it, like they go
back and they say, okay, this, you know, variant did not exist before this time
period. And then, you know, you find all these samples. And what you find
actually is that some hospital systems recycle ID numbers. And so after a
period, because they've gone around the counter, and then that means that
actually, you know, the metadata is associated with the wrong samples because
you're consistently updating the metadata for different systems. And so it's all
wrong. And you only notice it when you find variants that you absolutely know
didn't exist. Speaking of hospital IDs, speaking back to COVID as well, we had a
hospital ID which referred to the patient as opposed to how many times they've
been sampled and sequenced. And I was sort of just doing some sanity checks on
the data because I'd noticed that was only like two different case IDs, two
different hospital IDs in the data set. But they're all either blank or
3,200,000. And then I realized what had happened was that the hospital IDs are
10-digit numbers that start with 32 because it was 2020, something to do with
three is our state code or something. And Excel was converting them to
scientific form and rounding them. So any number 32001269 became 3.2E08. The
next patient was 3.2E08, 3.2E08. Everything was rounded and converted to
scientific form. And yeah, this kept happening over and over again.  In Excel,
it seems you cannot turn off this automatic conversion, as far as I can tell.
So, unless they manually went each time to the Excel spreadsheet and forced it
to be a string type, it would always get converted to exponential form. So, one
way I found out you can prefix cells in Excel with a single quote, and that will
treat it as a string literal and sort of avoid this, but when these things are
coming from other sources, you can't do that automatically. But then if you're
going to process that with a script, you don't want those single quotes in there
to mess things up. Well, I think when you export from Excel, a single quote's
not in there. It's used at the entry point to say that this is a literal, and
then when you export it back, it stays as a literal. But, yeah, messy. I mean,
we're all familiar with Excel and date formats, but this was one that I hadn't
really encountered before. So, always put a letter at the start of your ID codes
to avoid this problem, is my advice. Yeah, and we found that some hospital
systems, the same systems were sold to multiple different hospitals, and so it
wasn't globally unique, say, for the U.K., and so you'd have the same namespaces
effectively, which is a real, real pain, because then you'd have the hospital
identifiers, which we weren't allowed to have, because they're pseudonymized,
and then you had the patient, like the NHS numbers, which we weren't allowed to
have. So, we were using identifiers which were not globally unique, which is a
problem, and so then you had to go and combine it with patient age and data
collection and all these other things to try and make sure you had the right
patient. And on many occasions you had to re-upload data, because it would get
overwritten then by other systems, which would fix the metadata, and, yeah, it
was a bit of a pain, a huge pain. But then paperwork was another problem as
well. At one point, there was a change in how systems work, and instead of being
automated in APIs, it turned into, OK, for small sample numbers, up to about 20,
you'd have to send a piece of paper with each tube, and then that became a huge
time-consumer. If you have 20 pieces of paper for 20 samples, that's not
scalable at all, particularly when people are overwhelmed. So then those got
reversed. There's all these little changes that have downstream consequences
that people don't realize. Anyway, I'm going to stop before I get sued or
something. I think it was, I mean, obviously, by informatics, we were very busy
during the pandemic, but I think there were definitely times where I certainly
got utterly disconnected by quite how much labor at a certain stage was in the
lab process. Like at the height of the pandemic in the shared hospital lab, I
think we had something like 8 to 10 people full-time just putting caps back on
sample models, because they were coming off automatically on the machine, but
they had to be resealed for disposal or storage or whatever. There was so much
labor in that step, where it's like, oh, I just get the data that you have, and
the workflows do all the work. You get stuff, and you find that the quality of
data is really poor, and you say, oh, could you repeat that? And they're like,
no. There's no physical sample left. We dispose of them because we physically
don't have the freezer space to store the samples after a day. I mean, I was
taken by surprise by the, like I thought, okay, there's been a few QC failures
in this run that could do with a resequence. And over a certain number, it was
generally less effort to rerun the entire plate than rerun a small subset. And
just that scale of sequencing and economy was just a bit alien from the more
academic side, where you're trying to make, you're trying to scrape every
sequence you can out of your funding or whatever. Yeah, I know the cutoffs for
us were certainly very strict in terms of number of contaminated reads or number
of reads in the negative control. That was vastly stricter than we've ever done
in academia ever. It's the tiniest, tiniest bit, and we're just redoing a whole
plate, which is a lot of money. It's huge amount of time and effort. But now
we're finding that random coronavirus reads will just pop up everywhere in every
city, even though we haven't done anything since April. It's just a building.
Everything is just contaminated. All the reagents coming in have a tiny bit of
contamination. And you can tell because, say, if it's from internally, it's 400
base amplicons, and it's very clear that that plant does not have coronavirus,
particularly now that it's a 400 base amplicon. I think we'll be forever seeing
COVID contamination in all our data sets. Don't worry about all these very
sensitive, culture-independent data sets. With PTC in them. Yeah, we'll be
declaring every patient has died from COVID. Probably should reassure that these
are COVID amplicons, not functional viruses. These are just sequences.
Contamination floating around in the air. Yeah. Yeah, yeah. I'd say most labs
are fully contaminated with aerosolized COVID by now. Anyone who touched COVID,
like, yeah, it's just jam-packed everywhere. And all the companies that make the
positive control material is, you know, they're jam-packed COVID everywhere.
Turns out DNA is really small. Really? And common. It gets stuck in the nitty-
gritty corners of the lab. Yeah. There's not much you can do, you know, once
it's there. It's hard to get rid of. It's the new Salmonella. For us, Salmonella
was cropping up in everything because it was our most commonly sequenced thing.
Like, not at high levels, small levels, but it's there. But then you get people
doing these low biomass samples and finding microbiomes everywhere. It's like,
oh, the brain microbiome, you know, whatever. It's like, are you sure about that
one? Just like a lot of the big marine microbial eukaryote things, we often
would find different Malassezia species all over the world. And it's like, I
mean, yes, there are. It's a diverse group that might be everywhere. However,
your dandruff is also all over the world. But there was a long-term discussion
about the kind of everything is everywhere hypothesis being more related to the
fact that, you know, the people sampling were going everywhere. Yeah, yeah.
People licking their fingers or... Well, we've had that particular COVID, you
know, like the person collecting the sample, you know, often that was a problem,
you know, because, you know, you get CTs and maybe 37, 38. And you wonder, is
that real? Or is that just because they walk from one room to the next? You
know, like to start to sample people and maybe it's a tiny bit of COVID lying
around from positive patient. All these marginal cases. Anyway, does anyone else
have any other anecdotes? I was just, just keep thinking about dates and Excel.
Merging those two things together is just fraught. Yeah. I'm always getting
asked about a particular output that our tools generate. And the date apparently
is always wrong. But only on one person's computer. And because it's not my
computer, I can't reproduce the error. So if anybody has any insights into why
dates and Excel could be wrong, email me. Yeah, I believe dates are stored in
the Excel file in a kind of machine-independent form, like number of seconds
since 1900 or something like that. 1970, Unix epoch. Unix is 1970, but I thought
Windows used a different thing. And the setting on your computer describes what
date format to use. They've left their format as U.S. They don't always export
as U.S. Have they come from a country that uses a different date format to
Australia? Honestly, I don't know. And I've sat down and tried to figure it out
with them. And apparently every time they open it, it's wrong. Try it? Yeah, I
think look at their English dictionary settings and whether it's English
Australia. Or English U.S. It could be a version thing as well. Like there's
always quirky things going on. Yeah. But speaking of bad metadata, you just made
me think of Gazade. Be careful there, Naeem. Obviously validating metadata is
important, and Gazade doesn't validate metadata. And it's quite interesting. I
was just fascinated by the columns in Gazade looking at them. I was curating
them to clean them up for our own so we could compare our data to Gazade. And
the gender column, male and female, I realize that lots of cultures in the world
obviously don't use the English words for male and female, but they do start
with the letter M. Most of the words for male in— Irish is the opposite way
around. Menor is a woman, and fer is men. Well, the Irish used English
nomenclature, it seems, in Gazade mostly. But yeah, I couldn't believe how many
different languages were represented in the gender column. Also the age column
was sometimes written as 101 years, or it might be 13 months, or all sorts of
random things. So yeah, validate your metadata, people. The sequencers were
fascinating as well. And I don't think I never realized quite how many rebadged
sequencers were sold in different parts of the world. Really? I found, I think
it was in— I think it was Malaysia had clearly had— the IonTorrent had been
rebadged and sold under a different name by a different company that was in the
dataset. Emma Griffiths, when she was doing the metadata specification work, did
a lot of work looking at some of the weirder things that turned up in those
columns if you looked down at them and broke them down. No, I believe IonTorrent
has been resold and rebadged a lot in Southeast Asia, pushed into public health
labs as a rapid solution for amplicon sequencing. I just met someone recently
who wanted some help from a country in Southeast Asia, and yeah, they're using
IonTorrent for COVID. Very unusual. It was the sample prep, I think, they really
became very popular, because the IonChef sample-to-sequencer kind of automation
thing.  think was one of the reasons why it was as popular as it is. But I think
some private companies are selling it like a turnkey solution and rebadging it
as something else. So these people that are using that solution just call it
what the solution's called, not necessarily what the specific instrument is
called. That's mad. Managed ones have to, particularly an organization renowned
for being very high tech and the latest technologies, they'd have to spend a lot
of work persuading them not to do microarray in IonTorrent in the mid-2010s. So
our microarray is in storage. You know, we were building four years ago the
microarray into storage permanently. But yeah, people periodically say, oh,
well, we spent so much money on that, maybe we should pull it out. It's like, or
not. But going back to GISAID, I mean, the metadata is a bit crazy. You know,
often you have travelers and they're not counted as travelers in the database.
So you just see the country and then you don't know, you know, it's very, they
should annotate that this person is not from that country, but they're traveling
from somewhere else. To be fair, the people submitting the sequence may not be
able to access that metadata from the public health unit anyway. Whether it's
worse to have it there or we're missing metadata or not have it at all, I don't
know. Absolutely. But at least people were making their data public. I went and
reviewed loads of papers for COVID and I found that I couldn't reproduce most of
them, you know, a big chunk of them because there was no accession numbers or
links to the data. The data, say, raw reads weren't released. Maybe they just
released the consensus sequences or there wasn't metadata or there wasn't the
pipelines used to process the data. Everything's missing, you know, in one form
or another. So there's actually only a tiny number, maybe 30% of the papers were
actually looked reproducible. Now, I didn't know were they reproducible, but,
like, there's a big, big problem there. Well, and then the other challenge that
I ran into was versioning. I don't know whether it's been solved subsequently,
but versioning of sequences in GISAID. So if someone updated a sequence, the
accession didn't change. Really? So, yeah, you could have the same, you have
different sequences under the same accession. So it was all silently updated. I
don't know if they added a version more subsequently, but that was certainly an
issue last time I dug into it. Yeah, it's part of processing GISAID, so I
filtered it down. I encountered this duplication problem as well. The original
record would just stay in the database as well, so it was there twice. Yeah. I
mean, maybe GISAID isn't the best thing for this type of data. You know, maybe
you should focus on NCBI and EBI. But I mean, but I mean, it's one of the things
that, beyond, you know, arguments about data sovereignty, which are important
and always worth discussing, which GISAID obviously kind of spends a lot of time
leading the discussion thereof. But one of the driving factors, I think, behind
the use of GISAID was, you know, it's a very simple structure. You upload your
spreadsheet, you upload your sequences, it's there. The problem is that
structure doesn't represent the complexity of the actual data, which is what the
more complex structure of your bioproject, your biosample, your SRA, your
records are all linked together in the multiple different databases. It's way,
like, even with the best wizards, it's always going to be a more complex process
to integrate your data into because it's a better representation of the data,
and the data is that complex. And that's why you need to hire bioinformaticians,
computer scientists, and people to manage your big, important datasets. Keep us
employed forever. Michael Bentley podcast, maintaining jobs, securing and
providing for petitions. Yes, we need to have complex systems so that we can
have jobs. And no GUIs, no. That was Twitter the other week. Oh yeah, no GUIs. I
mean, GUIs, I agree in some cases, GUIs can help, like, say, if you're, say
you're a wetlaps artist, and you just want to do a bit of analysis, and I think
Galaxy's very good, and a lot of people in Quadromena are now using Galaxy for
analysis, and you have people who don't know the command line are able to go to
assemblies, they can find AMRDs and phone plasmids, and that's fantastic. They
can go in self-service. They don't need to use my team at all. But then,
obviously, there comes a point where you do need to do stuff on the command
line, which is most of what I do, to be honest. I'm always amazed at the usage
statistics for some of the online tool portals, like card RGI via that portal,
like, they've actually got some data that they're going to be releasing, like
the live usage, and it's wild how heavily it's used. I mean, I know myself,
like, if I need to do a little blast, like, I'll just use NCBI blast against NR.
It's just quick and simple, and I don't need to worry too much about, is the
database up to date? You know, do I have to do blah, blah, blah? It's just
straightforward to use. But there's definitely, and I think, again, even during
the pandemic, with things like Usher and stuff like that, definitely some public
health groups out there where using the web portal is the workflow, manually
taking the data, putting it in the web portal, copying down the data, like, that
is the workflow that this is being done using. I have found, like, some of the
workflows, sorry, some of the web pages were quite good, like Nextslade. I found
this fantastic, and I throw it in simply because you can take a screenshot of it
and say to people, listen, this is why I think your data is contaminated, or
this, you know, you can clearly see big blocks are missing here, and you can't
list this data. And, you know, it's a good way of visualizing and showing other
people where there's issues. And you don't have to write it. And I don't have to
write it either, yeah. Or if you just send someone a list of SNPs, you know,
it's meaningless to them. Or just if you condense it down into lineages, it's
not much use either, because you don't get, of course, subtleties of, you know,
within the data, you know. If important applicants are missing, then that
changes things a lot. Yeah, I think I've realized that us as bioinformaticians
who have access to high-performance computing and command line environments, and
we're the minority. Most of the world is generating fast, new data in their
labs, and these online web services and portals are like extremely highly used,
as Finley just said, with RGI. Obviously, the Danish Center for Genomic
Epidemiology, all their tools are massively used with millions of jobs run every
year. And yeah, I think that they have their place. And without them, a lot of
people wouldn't be able to do anything, so yeah. It's interesting the drive
towards, like a lot of these tools, like MaxClade, actually executing on the
local machine. And all the compute is actually being done locally, rather than
via the browsers. I did not know that. The browser, yeah, the browser's acting
as an execution engine. Oh, brilliant. Yeah, that's the next big thing, I think,
is client-side computation. And everyone's got a multi-core laptop now, and the
new web technologies, such as WebASM and WebWorkers and stuff, allow highly
parallel, fully, running on bare metal, essentially. The WASM is converted into
local, off-page, just-in-time conversion, really. It's not even compilation.
It's really a one-to-one conversion between WASM and the local, native
instruction set these days. So, yeah, there's gonna be, I think somebody already
demonstrated bacterial genome assembler running in a web browser. Seriously?
Will Rowe did a belt check for that, I think. He's gone into private industry.
Yeah, he has done. I was over at his house the other day. Client-side
computation is gonna be a big thing, I think, in the next decade in
bioinformatics. Yeah. And Rust, and all these up-and-coming languages can cross-
compile to WASM, so it's gonna be even easier and easier to get native code
running on a web client. It's amazing. So some of the challenges is the bugs can
be very hard to debug. If there's an issue with someone's, much like, even the
very personalized system, it's very hard to track down what's going wrong
sometimes with WASM. But, you know, in bioinformatics, it can take many days or
weeks to track down stuff as well. Yeah. Particularly if stuff isn't tested, or
written very well, or commented, or it's an obscure little, you know, bug.
Right, so we've covered dates. What else, what other kind of interesting, quirky
stories do you have? That we're allowed to talk about? That we're allowed to
talk about, you know, you might have to, you know, not name names. Well, I can
give you an old story. I don't know if I've mentioned this before on the podcast
series here, but many, many years ago, I went to a sort of a meeting, and it was
at the start of genomics when Roche 454 sequencing was all the rage. And I went
to this meeting, and one of the people was talking about, you know, their 454
run for their bacterial genome that they were working on, how they spent six
months curating their data. And I didn't quite understand what they were doing,
what they meant. And it turns out that they put all their 454 reads into a
spreadsheet, went for quality values in column two, and the sequence in column
one, and then manually looked at all the quality scores, and then in column
three, put a trimmed version of that read. And they would work on this every
Friday for six months until they managed to trim all the reads. Then they
exported those reads and used them for other analysis. And they still had
homopolymer errors. So I was dumbfounded. I just could not believe that this was
a high-level person as well. I was quite proud of their work. And as we all know
here sitting around this microphone that could have been done by a simple tool
in five minutes. And this person spent six months, one day a week of their life
manually trimming reads. But they looked busy. They did look busy. Yes. When the
boss was coming, they didn't have to switch to a fake spreadsheet because they
had a spreadsheet in front of them. And unfortunately, they couldn't get away
with that. You know, often people, if you look busy enough, you can get away
with it for a long time, you know? It's important work, if your manager doesn't.
how big or small the task is. I admire them for their persistence, it must have
been boring, especially after the second month. Or therapeutic, depending on
your perspective. Or maybe he went into a zen, or she, went into a zen mode. Or
maybe they did it for five minutes and then they wanted an excuse to not do any
work for a few months. But it was their own project. So it's just the slow
bioinformatics movement. Slow bioinformatics, I like it. Have you ever
encountered problems where, you know, a bioinformatician misinterprets something
because they don't understand biology, or that DNA has more than one strand?
Well, honey, you should say that. I was just thinking of a tool a while ago I
encountered. It was a tool for doing in silico PCR, and you'd give it a left
prime and a right prime and it would search your giant genome for that amplicon.
And I was running it and it was giving me results, but I couldn't figure out
what was going wrong, it wouldn't return all the results. And then it finally
hit me that, oh, maybe they're not checking the second strand. So I went back
and looked at the results, and yeah, all the coordinates they sent back were
always, you know, the first coordinate was smaller than the second one, so it
was only positive strands. So I emailed them and said, oh, I think you're
forgetting to check the second strand. And they said, oh, oh yeah. And they
eventually fixed it, but it just makes me wonder how many sort of these strand
bugs, and how many out by one errors. We all had the old, did we start counting
at one or did we start counting at zero, and these sorts of bedfalls versus GFF
coordinates and so forth. I wonder how many bugs are still out there related to
these problems. And you were telling me earlier that, you know, the start of a
gene can change. Yeah, I have to admit, as the author of Prokka, how
embarrassing it is to, when I first started in genomics, you know, I asked about
a gene and they told me about this start code on business and this end code on
business. And so I wrote this whole database, web database system for annotating
genomes manually and curating them. And then they let me know that start
positions can change, and my whole model was based on, like, unique keys
involving just the stop code on, and things like that. So basically I had to
live with this broken system for the next 10 years because it was too hard to
change. Yeah. Yeah, and I think the one for me was, I just thought all bacteria
had circular chromosomes and there was one chromosome in a bacteria. I never
realised that there's always exceptions to the rule, and so some bacteria are
not circular. Did you know that? And some bacteria can have more than one
chromosome. Well, I think you should say that the first bacteria I ever worked
on had two chromosomes. Vibrio? No, it was Leptospira, and yeah, it had one 3
megabase line and a 300 kb. And it was a genuine chromosome, it wasn't a
plasmid. Anyone else have any weird misconceptions about bacteria? I was, I was,
by doing the weird microbial eukaryote route first, I was well prepared for the
weirder, like, so the organism I did my PhD on, Paramecium brasserii and
alveoli, has two nuclei. It has a somatic nuclei and a germline nuclei. The
somatic nuclei is about 800 ploidy, and is basically expression profiled, so it
has all these intronic elements spliced out, and there's a bit of shuffling of
exons and stuff like that, you know, it's about Oxytrica does that as well. And
then it has this germline nucleus that's diploid and using sexual reproduction,
but is riddled with these intronic elements, like these invasive intronic
elements. So whether the second nucleus evolved because it became such a mess in
the germline, it's one question, but it doesn't bother with mitosis. Like by
doing, it doesn't bother chromosome segregation and mitosis, but this
macronucleus, it just randomly pinches somewhere near the middle, and the ploidy
is so high, it just about balances out the dosage. So like, and this is, this is
a, this is a serial, this has endosymbionts, it's a serial phagotrope, it's,
there's big, there's big DNA viruses in the system, like, it's a zoo, this
organism, a single set of organisms. So nothing bacteria has managed to throw at
me yet has stood up to the wildness of some of the microbial eukaryotes. I've
heard of some eukaryotes, like they'll have different ploidy at different points
in the life cycle. And I just blew my mind, you know, Jesus Christ, like we
assume everything is like human, or things like bacteria, but then you get all
the weird stuff that's out there in life, and it's just totally crazy. Like
every rule is broken. Wait till I tell you about plants, Andrew. Is it wheat
that doubles its genome every now and then, and it just, then it just has twice
as many chromosomes for a while, and then half, it jettisons half its genome at
random and continues on and just goes back and forth, I believe, yeah. I'm glad
I don't work on plants. But talking of bacteria-related, it reminds me of
Neisseria gonorrhea, which is interesting in that I believe that each cell has
between four and five copies of the genome in it. It doesn't have a single copy
of its genome, and I think the understanding is that this genome is pretty
clonal, but there is sometimes some variation, and that copies of the genome
within the cell can recombine with each other within the cell, intracellularly.
So it's fascinating, because an offshoot of this is that, so there's five copies
of the genome, and then each of these genomes has five copies of the 23S
ribosomal gene, and there's a well-known mutation in gono that confers
resistance to some particular antibiotic. It's a SNP in the 23S gene. So think
about it, we've got five copies of the genome, each has five copies of the 23,
so we have 25 copies of the 23S gene in this cell, and one or more of those 25
copies can have this SNP. If you have one copy of the SNP only, you're a little
bit resistant. If you have 25 copies, it's strongly resistant. There's this
linear dose response with the number of SNPs. So I don't think there's a tool
yet that kind of does a good job of measuring this particular SNP. So I know
that that's a thing that we've noticed is that, you know, we've had AMR tools
that predict genes, and now we're looking at AMR tools that predict SNPs, but we
need the tool to actually try and estimate the copy number of that SNP as a
proportion of the total cell, and we can't do that with an assembly. You have to
go back to the reads. And you need a specialist tool just for that species as
well. Well, it's a generic, it's a generic problem, right, finding a major
allele fraction in the data set, because you don't really need to know, you
don't need to untangle all these assemblies to do this. You just have to look at
the read depth on the 23S gene and kind of come up with a proportion. I don't
know if there is a specific tool. Maybe Areva can do it. I'm not sure. Christy,
could Areva tell you this sort of information? Areva could give you the fraction
of reads covering the 23S, but I don't think it would give you that in
proportion to the rest of the genome. It would just be in comparison. Well,
actually, as long as that's a proportion within the data set, that's enough to
know its potential. But doesn't Areva force one allele? Yeah. So if there's only
one step, that's okay. But it aligns the reads back. I think in its very
extensive output format, there is probably a column which tells you exactly what
you need, if you can find it. No, Areva is very, very well written in that
regard. That's a very common problem with 18S amplicon sequencing for microbial
eukaryotes because they often have multiple copies which have divergence from
one another. So how do you tell whether it's two different related species or
one genome where you have your amplicon sequence variants? It's the same
bacteria. If you look at 16S, often the intracromosome variation is greater than
the variation between the species. It's just crazy. Really? Yeah. For some of
them, we had some strep ones. What you do is you just take long read assemblies
and then you take the 16S and map them. You draw a file in a tree, then you see
what species they are. You can just visually see and it's like, oh yeah, okay,
that's totally wrong. That's wrong. That's wrong. Because you can't use 16S for
calling species. Does anybody in the wet lab know that? Genus is a stretch and
usually a wishy-washy. But yeah, no, it's a problem. Don't use 16S, there you
go. So thank you very much for joining me today, Kristi, Finlay, and Torsten.
Thanks for having us. 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 Microbinfee. 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.