Hello everyone thank you for coming to the NCBI webinar, on the evolution of NCBI-Hackathons.
Peter Cooper, the head of the communications group, is here with me to answer specific
questions.
I will answer more general questions at the end of the webinar.
Answers that we don't get to will be available linked to our webinars page.
Also the materials will be available on our FTP site.
All right let's go ahead and get started.
At this point, we have run 13 NCBI-Hackathons.
Over the course of the webinar, that tells you about what they are, what we do in these
NCBI style hackathons, and how you can get involved.
What we do in NCBI style hackathons in short, we put together teams of 5 to 6 individuals
that work on bioinformatics projects in a collaborative way, and 80 percent of those
teams finishes a functional software proto type at the end of three days,.
These are not three 24-hour days, in fact typically they are three nine-hour days.
Those nin hours are really filled with very focused work.
When we started these hackathons, we were teaching a lot of workshops in next generation
sequencing and we thought hackathons were a good way to offer advanced next generation
sequencing courses.
What we realized is what we were doing is we were rapidly building software prototypes,
ant that was the major motivations for people coming to these Hackathons, so what we ended
up doing was optimizing the Hackathons such that the teams develope software prototypes.
Like I sais earlier, about 80 percent of teams develope working software prototypes.
In our first hackathon we built four functional software products.
Including things like variant calling for RNA-Seq.
That was neat, and did a survey of available tools for variant calling for RNA-Seq.
Those tools have progressed a lot, but it was nice to have a place to see things in
one place.
In August 2015 we built six functional software products, and a couple of them including the
sort of maching learning Pharmacogenomics_Prediction_Pipeline, are still in active development.
In fact for the Hackathons in total, we've developed about 80 software products, and
about 40 are stable or still in active development.
This is a picture, a subset of the people that were in that particular Hackathon, and
this is approximately what an NCBI-Hackathon looks like, although as I will discuss later
they are now in many different locations.
All the tools that we build in NCBI hackathons are totally open source.
We put everything on github.
As I'll talk about that later, we also containerize things so we also see manifestations, on docker
hub, jupyter and other places as well.
We build several things in NCBI Hackathons, and things end up in several places.
Some things end up being worked in to NCBI production, I'd say that's a minority of things.
Other things end up being part of our standalone bioinformatics tools, but other things really
end up and being a part of the educational experiences and we work them into the webinars.
And also into larger scale online genomic courses.
One thing that I am particularly proud of, we built an on-line educational resource for
RNA-seq which is self-guided you can go to it right now and fire up some cloud instances
either on Google cloud engine or Amazon Web Services, and teach yourself how to do RNA-seq
mapping with STAR, HISAT and magic-BLAST.
Like I said, we worked this into an online workshop.
The first five lectures are available on YouTube right now.
In 2016 we extended the frequency, these hackathons have become really popular and we started
having them in other places, including Cold Spring Harbor at ISMB.
And also hack-seq was an NCBI style hackathon that was run in Vancouver and I was a participant
not an organizer of that event and it was a very special and very large NCBI style Hackathon.
So then in January we went back to our main Hackathon rotation.
We really started building things that integrated with larger software pipelines, including
things like SRA2R, which you can see here in the middle of yours screen, which is an
R package that will probably end up in Bioconductor as a way to port data from SRA into R frameworks.
We also started partnering with other groups.
I mentioned hackseq before.
We were also able to start building novel things.
Like for example, ways to compare structural variants to structural variant databases.
And that's been a neat focus of some of our work, working with structural variants, as
in my opinion there are not enough tools to analyze that kind of data in the Boinformatics
community.
Like I said, we started working with bioconda, with linux brew and with Bioconductor.
A lot of the things in our Hackathons, deal well with the sequence read archive, like
the SRA2R package that I mentioned earlier.
We are also interested in fostering comminity involvement with these hackathons and I'll
talk about that even more at the end.
But we were able to build a platform for metadata categorization for NLM indexers.
So they were able annotate data in GEO, which has been a bioinformatics community hobby
for the last 15 to 20 years.
They actually annotated all of the metadata for 14,000 Drosophila RNA-Seq datasets, which
we then used in a collaboration with FlyBase.
We have also been able to harmonize some of the metadata in DbGaP, with the NIH Common
Data Elements repository.
Another thing I am excited about is the ability to do really experimental things.
To our knowledge up to this point it was very difficult for people to put full genomes into
a graph, few people have attempted it.
So in a hackathon we ran in Cold Spring Harbor we were able to load seven full human genomes,
into a VGE phased graph structure.
That was something I think was quite neat, and a project that is ongoing.
We have also been able to work quite a bit on immunogenic peptides, looking at classifying
different types of immunogenic peptides from cancer, bacteria, as well as viruses.
This is something that I think is of particular clinical relevance, this is called DangerTrack.
This effectively is a public blacklist for regions of the human genome that are relatively
unstable and really unsuitable for algorithmic short variant calling.
This basically tells you where as a clinician you may want to double check before you report
a particular variant for a patient.
I have a question from the audience, asking if these hackathons are focused solely on
genomics.
This is a good question.
The hackathons are mostly genomics-based, but we have done some with some non-genomic-based
projects in some external hackathons.
In fact we had one project that was working one with EEG technology.
That said, the majority are based around genomics.
But with the big caveat that each January we have a biomedical informatics hackathon
at the National Library of Medicine.
And that really gets into a lot of things involving text mining and so on.
Thank you for the question.
In 2017 we have really evolved to emphasize containerization, publication and accessibility
with these Hackathon products.
And I'm really sort of proud of this because I think one of the things the hackathons contribute
to the community is speeding up science.
I told you before that at this point, about 80% of the Hackathon products become functional
prototypes, but another thing I'm particularly proud of is that about 10% of the get published
as manuscripts often within a couple of months of finishing the product in the Hackathon.
That to me is really quite neat.
I think accelerating publicaiton in my opinion is very important.
One of the things that is very accessible that we built in a hackathon is Phenvar.
And you can see the URL here is phenvar.colorado.edu.
This will generate word clouds of concepts co-mentioned with SNPs in PubMed abstracts.
But my friends like to remind me that word clouds are not really serious science, but
what is serious science is that is will also generate D3 rendered networks of SNPs and
in particular PMIDs.
That may be useful to some of the folks in the audience.
And to the question before, this is an example of some of the text mining stuff we do and
integrating this with genomics.
Something that is sort of pure text mining.
We built something called PubRunner.
I'm very excited about PubRunner.
Bascially what it is, this is a framework for narutral processing tools that continuously
updates the PubMed corpus.
This framework can also be used to update other corpora for natural language process
research, and we are working on packaging this and collaborating with a lot of other
groups, to allow them to auto update their PubMed corpus.
If you are interested in this, please send me an email.
It's ben, b-e-n dot b-u-s-b-y @nih.gov.
The head developer of this in the Hackathon, Jake Lever is currently working in the group
for the next four weeks on this project, so this would be a particularly good time to
contact me.
We have also been able to take other people's software products, and really put them into
more reproducible frameworks.
We have a common workflow language pipeline for epigenomics called SCREW.
And if you do epigenomcs, I'd really encourage you to check this out, because I think that
reproducibility is really important, particularly in the epigenomics space.
We have been able to collaborate, in this case with GA4GH.
This is a slide generated by the global alliance data working group.
It shows NCBI as a repository that works with their API.
At the time they made the slide, NCBI's data repositories actually did not work with their
reads API.
But in a Hackathon, we were able to stitch these two things together, and it really worked
quite beautifully.
And right now we are working to extend this to be able to work with raw data on the fly.
So the idea is the reads API can work even on unaligned data.
That is something I am particularly excited about as we go to more and more decentralized
databases for this type of genomic data.
In other things, we have done things like looking at resources for RNA-seq, both viewing
and doing counts, as well as identification of novel viruses.
This is something I'm particularly excited about.
But also when we announce teams that want to work on virus identification and integration,
they are extremely popular at the Hackathon's.
So I think this is a topic that really grabs the imagination and enthusiasm of folks that
attend our hackathons.
Speaking of, I want to take a brief segue and talk about the kind of people that come
to our hackathons.
Typically they are computational biologicals who are either senior graduates or postdocs.
And people who lead teams at hackathons are typically senior postdocs or people at the
assistant professor level.
There are exceptions to every rule we have had fairly young people attend and even lead
teams at hackathons.
If you are interested please contact me directly or simply sign up for a Hackathon.
We have been able to foster community involvement with the Hackathon's projects in a couple
of other ways.
We are starting to build educational resources that help people leverage community tools.
And I will talk about an expansion of this later.
This is the EDirect Cookbook, which will be mentioned in next week's NCBI webinar on API's.
But, I'll show you that these are jupyter notebooks that we are building to enhance
NCBI education.
We are hoping to build a whole suite of computational resources that are used to leverage NCBI databases.
We think that this will be helpful to the community.
If you are interested in getting involved in those projects whether through Hackathons
or through other means, please fill free to contact myself or Peter Cooper.
Next I want to show you some cutting edge stuff that we built just a couple of weeks
ago, these are projects I am very excited about.
I'm proud to say they are still fairly stable and in ongoing development.
One of the things that we built is a simple, very easy to use tool for looking at antimicrobial
resistance genes in metagenomic data sets.
This heavily leverages Magic-BLAST, which is a new flavor of BLAST optimized to deal
with next generation sequencing reads that we built at NCBI, and also uses the CARD database
to identify some of these metagenomic resistance genes.
Although I think it is attractive to integrate databases as well as put in some of learning
modules.
I encourage you to go to github, grab some code.
Check it out.
It's docklerized and very easy to leverage and use on anybody's linux system.
Right now the cancer genomics cloud is free.
You can also leverage it there.
That may be something to check out.
We are also interested in streaming data.
What this particular thing does, we built a database of adapters, that are present in
the SRA.
If you know what the adapters are, in SRA, for the particular dataset, then you can trim
them on the fly.
If you want to do assembly, and generate contigs, now you do not have to do fastq-dump anymore,
you can simply trim your adaptors on the fly, and then build contigs from streaming data,
which is something I think is very attractive to the NCBI infrastructure; also, with genomic
data getting larger.
If you have any general questions on that I would love to answer them; you can email
me directly.
I was particularly proud of a team, coming from and a particular academic institution,
and they wanted to build a viewer for an excellent mouse aging dataset.
What they ended up doing was building modulular types of viewers for existing RNA-seq viewers.
I think that is important because there are currently 35 RNA-seq viewers that are up in
the community and supported; about 35.
I think it is really important to say, if you have a particular thing you want to be
able to view, it is great to build modular tools that you can insert into existing viewers.
This is something that we have become interested in.
At the end of the Hackathon, we are able to generate movies looking at mouse aging with
and without infections to flu.
And what is really amazing to me, Even at this extremely zoomed out level of the mouse
genome, you are able to see responses to influenza, over long periods of time.
There is a question about virtual Hackathons but I will wait until the end to get into
that.
This was a really experimental project.
I look forward to iterating on this project, to be frank, this team was filled with brilliant
people, but I'm not sure random flipping is the most prudent obfuscation model.
What this does is send a robot into anybody's data set, looks for particular variants, then
if it is desired by the people who own the data, it obfuscates the data so it is no longer
personally identifiable.
That switch can be flipped back off to generate a full data matrix if some collaboration agreement
is reached.
Obviously consistent with the patient consents on the data.
We have also been helping out with other people's Hackathon's.
That's something I'm particularly proud of.
We have helped out with Bio-IT, as well as recently, an artificial intelligence genomics
hackathon in San Fransisco.
And these are things we look forward to working with.
Also on the NCBI Hackathons github site, which I'll put up on the last slide, you can find
a repo for being able to run your own NCBI style Hackathons, which I would be happy to
help out with as well.
But it you just want to roll out your own please feel free to take any pieces of this
and do that.
We would like to see more of this, there is no shortage of folks interested.
Our next hackathon is in Pittsburg PA at the end of September.
We will likely work on an out-of-the-box automatic Hackathon builder.
I am pretty excited about that.
I mentioned publications out of the hackathon, and one of the places we publish these hackathon
results, and allow others to publish these results is in F1000 research and they have
been wonderful in working with us in publishing Hackathon tools.
But certainly some communication of these tools has gone to other journals as well.
You can also follow @DCgenomics on twitter, and other places on twitter we try to advertise
things in these hackathon.
This is a picture of some of the folks in the most recent New York Genome Center Hackathon,
in June.
It was a fantastic experience.
We had a lot of folks from that metro area as well as folks that came in from across
the country and other countries.
Like I said here is a list of some of the upcoming Hackathons.
If any of you are in California, there will almost certainly be a Northern California
hackathon the first week in April, likely at either UCSC or UCSF.
We're trying to work on a Southern California Hackathon that may occur in January.
Please check this site for that.
Finally you can find the actual repos both at this github organization, and stable and
other projects that are continuing to be developed can be found at this URL, ncbi-hackatons.github.io.
Thank you very much for your time.
We have blog posts coming out on NCBI Insights about some of the products from the last three
hackathons; Boulder, CO, NY Genome Center, as well as the one we just ran at NCBI.
Now I'm going to take questions.
I actually have one queued up here.
This is an excellent question.
Are there plans to host virtual Hackathons?
We have done quite a few experiments with virtual hackathons.
One thing we have seen is that when we involve virtual people, we actually see not even sort
of less production, but dramatically less production from the people that are involved
online.
That has been our experience so far.
However, the last experiment we did on that, we had Hackathon veteran.
Someone who had been to several hackathons working part-time to work on modular tools,
and that was a very positive experience.
We've also had people who did quite a bit of work done online after the hackathons.
What I can see a model of is Hackathon veterans, or people coming to in person Hackathons of
this style, and then continuing to work on them independently.
That said, of course there are other models that work well online, sort of long-term competitive
hackathons tend to work very well online.
One of the things we've thought about doing is in the virus discovery space, as well as
antimicrobial resistance space, is hosting long-term online Hackathons to leverage some
of the tools we have built in the in person hackathons and use them to analyze data sets.
I would also like to mention there will be a second year of hackseq.
That's the big, awesome genomics hackathon in Vancouver I mentioned.
You can go there and register at hackseq.com.
A neat thing is that now there are other hackathons running on this model.
Someone else asked, does the hackathon have a Slack channel or team?
In fact this is a great question.
Each hackathon has it's own Slack organization, which obviously persist after the hackathons.
We originally ran them through Google groups and than through Slack and that's been very
popular.
We also have a linkedIn in group for hackathon alumni, where jobs are posted, people can
talk about issues, that sort of thing, and we also make announcements.
That's one thing I am very proud of, what I think we are doing with the NCBI Hackathons,
and other NCBI style hackathons like hackeq, is we are creating a community of excellent
computational biologists, but are also seeing evidence to suggest these people work well
with other people.
And really in my experience, they are really kind and nice people.
I think this actually makes an excellent tool for building a phenomenal computational biology
workforce.
That alumni group may expand from being a mostly linkedIn presence, to Slack and other
avenues of social media.
I would like to mention in the third week of June 2018, we are having Hackathon in Boulder
Colorado, as a part of a genomics conference, and we would like that to be what we are calling,
mostly a champions Hackathon.
So, mostly hackathon alumni that come to Boulder and work to integrate some of the projects,
and to build larger scale bioinformatics platforms.
Bringing those people back and having them continue to work for the bioinformatics community.
Finally we are starting to integrate with other bioinformatics hackathons that you may
be familiar with.
I have been to the BOSC Hackathons and we're talking with them about integrating, as well
as the general biohackathons community, and elixir.
Those are things I look forward to telling you about sometime soon.
Great.
Thank you for the really great questions, people who have asked questions.
With that, typically many times I don't put the Q&A session on the recording, but I think
perhaps for this one it will be popular.
Thank you to everyone who attended.
All of you Hackathon alumni out there; there's a couple in attendance here.
With that we will sign off.
If you have any further questions please feel free to email me.
We just have one or two spots available for the Pittsburgh Hackathon.
The deadline is closed, but if you are able to go to Pittsburgh September 25, 26 and 27,
please go ahead and sign up.
Thank you very much everybody have a fantastic rest of your day and a fantastic week tune
in next week for an NCBI webinar on NCBI APIs, you will find there are a lot more NCBI APIs
you thought there were.
For more infomation >> How to draw Mart register for Baby Coloring pages and drawing toys cash register for kids colours - Duration: 4:07. 
For more infomation >> Homekeepers - Carol Kent, Help for Prisoners and Their Families! - Duration: 28:30. 
Không có nhận xét nào:
Đăng nhận xét