.huge[Real-time Genomic Sequencing]

# .huge[Real-time Genomic Sequencing]
## Assessing the potential of <del>autonomous AI devices</del> <a href="https://www.nvidia.com/en-us/autonomous-machines/embedded-systems/">cool tech</a> for portable real-time sequencing
### <div class="line-block"><a href="http://sirselim.github.io/">Miles Benton</a><br />
Senior Scientist Bioinformatics<br />
Human Genomics, Institute of Environmental Science and Research (ESR)<br />
<br />
.large[<a href="https://www.eresearchnz2020.org.nz/">eResearch</a>, Dunedin, 12<small><sup>th</sup></small> - 14<small><sup>th</sup></small> Feburary 2020]<br />
<br />
<br />
.center[<img src="images/ESR_SFC_logo.png" style="width: 25%; border: 2px solid white;"/>]</div>

---

# .center[Advocate for reproducible research...]

## .center[Where possible my presentations and code are available online]

<br />

<p>
.center[
<img src="images/github_logo.png" style="width: 520px; margin-right: 1%; margin-top: 1.5em;"/>
<img src="images/sirselim_qrcode.png" style="width: 202px; margin-right: 1%; margin-top: 1.5em;"/>
]
</p>

???

These are my presenter notes. :)

The theme of eResearch 2020 is "**United in Data**".

These links throughout are a combination of github and hackMD, all should be available.

---
class: middle

[github.com/GenomicsAotearoa](https://github.com/GenomicsAotearoa)
]]

???

Part of the Bioinformatics Leadership Team, for health.

---
class: middle

# GPU basecalling (live demo part 1)

<br>

data: mixed bacterial sample, ~0.5 Mb (or ~5.5 Gb actual data)

```bash
# fast basecalling mode
guppy_basecaller \
  --disable_pings \                            # don't call home
  --compress_fastq \
  -c dna_r9.4.1_450bps_fast.cfg \              # model file
  -i fast5/ \                                  # input dir (fast5 files)
  -s flongle_test \                            # output dir
  -x 'auto' \                                  # GPU configuration
  --recursive \
  --num_callers 4 \
  --gpu_runners_per_device 8 \
  --chunks_per_runner 256
```

.pull.right[<span style="color:#3498DB">... **que to start ~~trial by fire~~ the first demo** ...</span>]

???

Notes

---
class: middle

# while that's running...

???

Refer to Peter's opening address, reiterating that genomics is important and increasingly so.

Check to see audience knowledge of genetics/genomics.

A little genomics 101 if needed...

---
layout: false
class: middle inverse

background-image: url("https://cdn.vox-cdn.com/thumbor/llQlREwACaitewdPcLm5HzWT_g0=/0x1:1100x734/920x613/filters:focal(0x1:1100x734):format(webp)/cdn.vox-cdn.com/imported_assets/1507663/DNA-sequence.jpg")
background-size: cover

???

DNA - A, T, C, G
Part of all 'living' things

---
class: top

# what are we doing?

* <span style="color:#3498DB">**the idea**</span>

* low cost
  * accessible
  * portable
  * fun!

--
  
* <span style="color:#3498DB">**example use cases**</span>

* field sequencing (real-time monitoring, forensics, agriculture, waterways, ...)
  * **clinical settings**
  * community outreach / teaching

<p>
.center[
<img src="https://www.illumina.com/content/dam/illumina-marketing/images/decade-in-sequencing/decade-in-sequencing-320-web-graphic.jpg" style="width: 820px; margin-right: 1%; margin-top: 1.5em;"/>
]
</p>

.small[.center[
(image source: [illumina.com](https://www.illumina.com/techniques/sequencing/dna-sequencing.html))
]]

???

I used the word portable in my previous slide, this is not a phrase one would attribute to machines from Illumina / PacBio.

These things are huge! ... and expensive.

---
layout: false
class: middle

<p>
.center[
<img src="https://cdn-a.william-reed.com/var/wrbm_gb_food_pharma/storage/images/publications/food-beverage-nutrition/foodnavigator.com/article/2018/03/21/oxford-nanopore-raises-funds-to-support-commercial-expansion/7992741-1-eng-GB/Oxford-Nanopore-raises-funds-to-support-commercial-expansion_wrbm_large.jpg" style="width: 620px; margin-right: 1%; margin-top: 1.5em;"/>
]
</p>

.small[.center[
(image source: [nanoporetech](https://nanoporetech.com/about-us/news/oxford-nanopore-announces-ps100-million-140m-fundraising-global-investors))
]]

???

flongle: ~$500 USD for 5
1 might be enough for a metagenome

minION: $1000

smidgION: still just a concept

gridION: ~$50K

promethION: ~$230K

---
layout: false

<a href="https://f1000research.com/slides/8-1947" style="position:absolute; top:625px; left:190px">slide kindly supplied by David Eccles (f1000 presentation)</a>
<a href="https://www.youtube.com/watch?v=CHCAb-PAqUI" style="position:absolute; top:652px; left:190px">"Sequencing DNA with Linux Cores and Naopores"</a>

???

David's talk and a live demo are linked.

---
layout: false
class: middle

# Example squiggle plot

![](https://nanoporetech.github.io/tombo/_images/single_samp.png)

.small[.center[
(image source: [tombo manual](https://nanoporetech.github.io/tombo/plotting.html))
]]

???

What we 'see' when DNA passes through the pores.

---
class: middle

# Why GPUs?

* squiggle data lends itself nicely to neural nets
* GPUs are very capable in this space
  - CUDA cores

ESR GPU basecalling benchmarks [(link)](https://esr-nz.github.io/gpu_basecalling_testing/gpu_benchmarking.html)
* Titan RTX & 2x Tesla V100
]

<br>

---
class: middle

# Our experiences with the Xavier [(link)](https://hackmd.io/@Miles/HkumH7sBH)

.pull-left[.right[
<img src="images/IMG_20191021_165639.jpg" style="width: 200px;"/>
<img src="images/IMG_20191025_094614.jpg" style="width: 200px;"/>
]]

.pull-right[.left[
<img src="images/battery_charge.jpg" style="width: 200px;"/>
<img src="images/xavier_in_use.jpg" style="width: 200px;"/>
]]

???

Talk about the Xavier specs:

* 8 core arm 
 * 16 GB of RAM
 * 512 CUDA core and 64 tensor cores
 * 512 GB SSD NVMe

---
class: top

# Benchmarking [(link)<sup>*</sup>](https://gist.github.com/sirselim/2ebe2807112fae93809aa18f096dbb94)

<br>

Most Jetson devices have the ability to be put into different power modes

<br>

<div id="htmlwidget-737dc01837087d5cd947" style="width:100%;height:auto;" class="datatables html-widget"></div>
<script type="application/json" data-for="htmlwidget-737dc01837087d5cd947">{"x":{"filter":"none","fillContainer":false,"data":[["10W","15W","30W 2 cores","30W 4 cores","30W MAX (8 cores)"],["33.4 mins","14.3 mins","10.8 mins","10.8 mins","7.5 mins"]],"container":"<table class=\"display\">\n  <thead>\n    <tr>\n      <th>Power.mode<\/th>\n      <th>Time<\/th>\n    <\/tr>\n  <\/thead>\n<\/table>","options":{"dom":"t","order":[],"autoWidth":false,"orderClasses":false}},"evals":[],"jsHooks":[]}</script>
.small[
***Note:*** this table represents basecalling performed when running on a portable powerbank.
]

???

Why is 10W mode exciting?

We're running in 30W MAX mode for this live demo - running on a portable powerpack.

---
class: top

## Very rough comparison of Nvidia ecosystem<sup>#</sup>

<div id="htmlwidget-fee71ac527eb41663326" style="width:100%;height:auto;" class="datatables html-widget"></div>
<script type="application/json" data-for="htmlwidget-fee71ac527eb41663326">{"x":{"filter":"none","fillContainer":false,"data":[["Jetson Nano","Jetson Xavier NX*","Jetson Xavier","Geforce 1080 Ti","Geforce 2080 Ti","Titan RTX","Tesla V100"],["~0.5","6","11","11.3","~13-14","16.3","16.4"],["$99","$399","$699","~$650","~$800-$1000","~$3000","~$8500"],["5-10W","10-15W","10-30W","215W","250W","280W","250W"]],"container":"<table class=\"display\">\n  <thead>\n    <tr>\n      <th>Card<\/th>\n      <th>TFLOPS<\/th>\n      <th>Price<\/th>\n      <th>Power<\/th>\n    <\/tr>\n  <\/thead>\n<\/table>","options":{"dom":"t","order":[],"autoWidth":false,"orderClasses":false}},"evals":[],"jsHooks":[]}</script>

.small[Price in USD]  
.small[<sup>#</sup> a very small selection of what Nvidia offers obviously!]  
.small[\* [Available March 2020](https://developer.nvidia.com/embedded/jetson-xavier-nx), this can fit in an existing Jetson Nano board]

???

V100 could rip through this data in a couple of mins, but consider the cost, power, etc.

Xavier NX is exciting, module that can slot into the Nano body [**hold up Nano device**]

---
class: middle

# Results...?

<br>

.pull.right[
<span style="color:#3498DB">... **que to open the terminal** ...</span>
]

???

Show completed run, time and other metrics.

Also open/show head of a fastq file (cat it and take a long read to blast)

---
class: middle

# Case study (live demo part 2)

## Clinical metagenomics

<br>

.pull.right[<span style="color:#3498DB">... **que to start the second demo** ...</span>]

---
class: middle

<p>
.center[
<img src="images/jamie_sm.jpg" style="width: 330px; margin-right: 1%; margin-top: 1.5em; border: 3px solid white;"/>
<img src="images/jamie_big.jpg" style="width: 334px; margin-right: 1%; margin-top: 1.5em; border: 3px solid white;"/>
]
</p>

---
class: top

# case study

* had a lumbar puncture ('9 days' old) 
  * (after being born 8 weeks early)

<br>

* suspected meningitis (bacteria/viral)?
 * told after the fact

<br>

* \>**48 hours** for results return

]

---
class: top

# case study

<br>

.large[
* a sample (i.e. blood) taken,
* DNA/RNA extracted/enriched,
* sequencing library prepared,
* nanopore sequencing initiated,
* sequence starts coming in almost instantly,
* **<span style="color:#3498DB">time to first read-match detection <1 hr</span>**

]

<br>

???

Stress how fast and low cost this has the potential to be.

---
class: middle

# Results...?

<br>

.center[
## BLAST [blastn](https://blast.ncbi.nlm.nih.gov/Blast.cgi?PROGRAM=blastn&PAGE_TYPE=BlastSearch&LINK_LOC=blasthome)
]

```bash
zcat flongle_test/fastq_runid_3b07b3e214d5928e9ae59bd18918e7298d4f601b_20_0.fastq.gz | \
  paste - - - - | \
  awk -F '\t' '{L=length($2);if(L>M) {M=L;R=$0;}} END {print R;}' | \
  tr "\t" "\n" | \
  sed -n 2p
```

???

Take an example read and blast it.

Quick and dirty oneliner to find the longest read in a fastq file, we can then paste this into our blast window and hopefully identify a speices of bacteria...

---
class: middle

---
class: middle

### What's actually in there?

<br>

Mixed sample of bacterial species:

> 18AR0083 Enterobacter cloacae  
> 18AR0155 Klebsiella pneumoniae  
> 18AR0156 Escherichia coli  
> 18AR0164 Escherichia coli

]

???

Success!

---
class: middle

# Summary ...

<br>

## ... if things haven't crashed and burned ...

???

Again reiterate:

* cost
* portability
* accessible - off the shelf parts
* fast
* GPU tech scaled down
* MANY use cases

---
class: top

# acknowledgements (people to 'nerd' out with!)

<div id="images">
    <a href="http://xyz.com/hello">
        <img src="images/RDean.jpg" width="350px">
        <div class="caption"><span style="color:#3498DB">Richard Dean (ESR)</span></div>
    </a>
    <a href="http://xyz.com/hello">
        <img src="images/DC_ESR_2.jpg" width="350px">
        <div class="caption"><span style="color:#3498DB">Joep de Ligt (ESR)</span></div>
    </a>
</div>

David Eccles (Malaghan, Gringene Bioinformatics)
]]