HPC by Industry: Bio

indeed yes enjoying it quite a bit out here hi everyone I'm Forrest Burt I'm an HPC systems engineer here at ciq excellent how about now Alan yes uh sorry browsers are weird things um yeah Alan still I'm managing director of the high performance Computing Center at Texas Tech University and uh one of several uh co-directors of the multi-university National Science foundation's cloud and autonomic Computing industry University Cooperative Research Center which is a mouthful hi thank you so at any point in time if you guys have questions go ahead and post those in the chat we're going to go ahead and start off and I know

bioscience is a very broad field and I know that there are some key areas that that HPC really plays into that what do you guys see as those types of fields and what do those applications look like and I'm going to pick on Glenn first because he's in the middle and he's still on mute let's wait all right Brock go ahead well uh I I think it's very very broad um you know drug design is one area you know I would say it's molecular Dynamics in drug design um obviously Glenn if if you were able to say it would probably jump in with genomics or

bioinformatics um and then parts of you know what I would love to get to talk about today is as well there's a you know a lot of crossover into um things that may be more like simulation mechanical engineering or something you know there's there's medical devices and things that have to go through similar simulations and um you know testing to be incorporated into the medical fields and I saw a presentation a couple months back I don't want to say exactly what it was but it was a device that was implanted into the body and obviously something that um needs to have a very Defined Life

lifetime of operation as well as durability because if it fails you potentially could be fatal to the the patient right so those there's simulations and things going on and that's that's a little bit of a crossover that's not necessarily and thought of as biosciences but it's related to biosciences field and high performance Computing is very key to some of those those testing so few areas yeah I'm gonna come back and ask you a question specifically about that in a minute now that Glenn appears to be off mute if we can hear him as a whole different story but of course so um yeah so Brock

uh got us started off really well uh molecular dynamics of course uh comes into play both in kind of like uh basic research you know looking at the way proteins fold and uh particularly in drug design how drugs dock to different protein receptors uh what I've spent most of my time on in the space and Brock also mentioned was bioinformatics that that tends to be um involved in a couple different fields one is like predictive Health analytics right so you take some bioinformatics genomic data and you take some electronic health records and you try to do some modeling and predict um you know how how

often someone uh or what are the chances someone would have to come back into the ER with sepsis after treatment for example um Precision medicine is really kind of where bioinformatics and genomics has kind of pointed to so if you if someone were to have uh be diagnosed with the cancer and we were able to sequence their genome and compare it to their cement normal cells we could try to pick out those mutations in the in the tumor and pick drugs that would specifically work for the that patient's mutations their background uh and not rely on just any old clinical trial that's been done in

the past um so uh and then of course AI um with with regard to let's just talk about the genome for a second if I want to uh take all the human genomes I can get my hands on for example and I want to look for um how does how does DNA replication start uh what sites do they start at well I know the basic pattern um but I want to expand on that so I can try to you know build models and train with all sorts of data looking for like new sites for DNA replication or different types of sites things like that so

we can learn things about the genome as well um and uh last but not least uh modeling disease spread like pandemics for example um uh there was you know obviously a lot of that done with um with the recent or current I should say code pandemic um but it's been done it's it's done it's being done right now right for the outbreak of Ebola in Uganda right so uh it happens every time you know there's there's a uh there's an outbreak of things like that so that's uh and those those simulations take in massive amounts of data right you got environmental you've got traffic you've

got weather patterns you've got air traffic you've got uh other geological environmental factors to be built in so it's not necessarily just a compute uh problem but genomics and modeling of these pandemics are just massive data problems excellent John this is a a place that you spend most of your time seems like yeah I mean it the elf covered pretty much the wide gamut of stuff that we do in bioscience so the only thing I could add to that is when you think of the word bioscience biophysics biochemistry biogeology bio is a prefix like bio gives us a context and a reason for doing all

that other science so you know in my view and it's uh clouded Maybe by having a couple of biology degrees it is the most important science period thank you Gary what are you seeing here let me just add to the list uh we have a lot of large instruments at the at the laboratory and uh so there are areas that require Imaging that require the HPC systems to reconstruct the image so some of the things the pipelines I'm thinking of are like cry OEM and typography and stuff like that thank you Dr Steele hey so um yeah only my mother calls me that um but

it's fun I don't blame that on purpose so yeah uh so this is a huge range of things already covered in the comments and I agree with all of them um I can't resist um pushing back on our Chan Zuckerberg colleague here a bit um uh I will set up an immediately demolish an old meme for physicists I'm a particle physicist by background and there's uh uh Lord Kelvin once said all all signs either physics or stamp collecting but I'll immediately demolish that because that was maybe true when I was a high school student but uh since then the information science of biology has come

along that's what we basically are talking about here and it's astonishing the light years of progress in just a few years of work so um I guess what I want to add to the previous comments is the way in which we approach biology topics as in the information Sciences uh and even there there's a huge range of tools that that just dwarfs anything that was available even just a decade ago so everything and we've discussed many of these before on this this broadcast um everything from in browser apps where you can run you know uh python or Jupiter notebooks everything uh things that used to

require a whole cluster to do that you can just do without leaving your browser even holy cow and uh all the way up to things that occupy the biggest clusters so uh and Beyond single machines or clusters there's there's a collaborative aspects of the use of technology in biology a few years ago the NIH started a cloud project um and if you've heard previous broadcasts you know that we've discussed the economics and the pluses and bonuses of cloud versus on-premises and why many big scientific outfits still run on premises resources before but NIH did something which uh required I think the cloud to resolve which

was to made data sharing among biological sciences practical and efficient so there's all these dimensions in addition to the science itself I'll stop there no it's great and I there's one theme that keeps coming up that I'm gonna go back to Glennon in just a minute but Jonathan Forrest do you have anything that we've missed couldn't hope to find something that's been missed by this so far all right I I actually uh I want to jump back in well of course you you go on if you have some I I was just really quickly I didn't hear anyone say it by name but some of

the advances that have been powered by HBC and biosciences are things like Alpha fold which you know took huge amounts of data about protein folding and was able to then predict um you know I've always heard decades worth of protein folding research within just you know the matter of weeks that it took to utilize the AI model to generate all of it um so HPC is huge in biological sciences and we're really seeing amazing advancements um powered by HBC especially around Ai and stuff in the field at the moment but I didn't hear anyone say that use case specifically but wanted to mention that thank

you for it and that's something something that Alan said that um made me remember you know drug trials today you know the FDA still mandates uh human trials you have to actually have those trials and we're moving to an age where we've got data now where you can simulate populations of people and that can speed up the testing of drugs it also can actually reduce the the points where you're having to do things like double blinds where you know literally some people who could be helped by drug uh don't get it because you've got to see you know what the impact is and so there's

there's a lot to unpack there when you're talking about using simulation for the for FDA approvals do you trust that the simulation output is correct um the the ethical element of are the models trained for enough populations you know so you've we've got to consider a lot of different things in in terms of using that data in that technology but it could really change the game in terms of getting you know targeted drugs to Market faster that you know again you look at the covid-19 vaccines and and the short amount of time that we have and you have to make these really tough calls do

you push a drug out there without full knowledge of what it can do so AI HPC you know the underlying computational Technologies I think have a way that can really help impact the health of of you know the planet people worldwide so can I ask a question about a gap that I don't think we've eliminated yet which is the gap between many um practitioners and biological sciences and the Computing Technologies required I mean uh many of the research researchers that use our systems uh just admittedly just self-admittedly say they haven't a clue how to get started whereas others are really sophisticated so um I'm sure

you've seen this I figured Brock would have an opinion on that one but Gary is that something I think we've kind of touched on this before but is that does that seem to be the case of what you're seeing is there the people who are good at it and the people who are bad at it in the middle just kind of seems to not really be the middle yeah you know I I think the um the the people in this is just a generalization so it's not doesn't apply to everybody but I I think as a whole the people in life sciences uh have used

computation you know are more recent to using computation uh than maybe some of the other scientific disciplines and so um and uh and I'm just seeing it like a new generation of researchers who really don't want to even understand some of the underlying technology and say code uh so that that's becoming uh something that I'm seeing um and then uh the last thing I you know I think which is kind of an implementation barrier is that uh when researchers are working with sensitive data you know they it's one thing to like secure a server or their own workstation you know to work on some of

this very sensitive data but then um you know to move it to something that requires high performance Computing means that you have to apply those security controls to a research cluster which is a huge undertaking and a lot of Institutions you know that's a big step that's hard to implement because a lot of the Cyber Security Experts really haven't had that experience of applying the controls to you know what we do every day with clusters and workflows and stuff thank you Gary yeah John I I was thinking when we were talking about this I was thinking about uh something that happened to me when I

was an undergraduate I was sitting outside a Dean's office waiting to talk to my turn to talk to an advisor and the person ahead of me that was sitting in the advisor's office said you know I really like science but I'm not good at math and the advisor without any hesitation said oh biology you should go into biology and I cringed then and I cringe now but a lot of the problems we face with educating users on how to use these systems stems from that attitude towards biology that I think is still prevalent in the educational system um you people who go into biology think

they're not going to be doing math and they think they're not going to be doing computing and that's not true anymore like you you don't do biology anymore without doing math Computing I mean there are a few people out there who maybe just go out in the field and describe you know new species or mating habits of a beetle or something but those as a biologist go those are becoming really really rare and I don't think the biology coursework across the educational system has caught up to that it's interesting and I don't guess I've ever really thought about that before but now it makes me

wonder if Glenn's a mathlete and I didn't know it I am I am not a mathlete I uh John brings up a really good point uh because in in at U of I University of Illinois there were kind of two tracks there was one for like the engineers and there's one for the biologists and so I had my own track of you know calculus and whatever came after calculus I did get into differential equations but like the engineers had a different you know physical chemistry course and not you know versus mine and so mine dealt with you know uh lipids and uh and uh micelles

and stuff like that and the engineers course you know dealt with you know just uh surfaces and materials so it was a really clear difference there but we're all using you know pencil and paper but now like John said and like unless at minimum you're using Excel and a laptop in the field right but you still got all sorts of other data uh if you're studying your frogs or fungus on frogs you've got data from people coming worldwide that you have to compare it to so not only do you have to you should understand the statistics that are being run into your spreadsheet but now

you want to you know compare it and normalize it to you know different data sets worldwide so you really just don't have that option anymore uh to kind of you know not be at least understand you know the math that you're working with and you know Gary you know I I've been forced to grudgingly accept that there are biologists that just don't want to understand uh kind of what's going on behind the scenes of their Pipeline and they just want browsers because I I thought that you know once we started minting bioinformaticians with actual bioinformatics degrees out of I think some of the first places

were like Santa Cruz and Stanford that that would catch on more and more and more and more people would have that training but it's actually just turned out to be the opposite you know things like Galaxy and Jupiter notebook are just rampant which is fine as a tool but so you know not wanting to grasp what's behind it and you know if you're going to be making a diagnostic call on a tumor right like I I have a real problem with that um I've come to I've come to grip that like that that's gonna continue and I'm still I'm still fighting against it but um

that's actually I I dealt that a lot uh with a lot of that t-gen it's like the bar for magicians kind of knew what's up and they created a completely different set of problems on the cluster and the other users just where you just help them try to figure out how to use a cluster so um everything seemed to fall in those two those two cams I know you've been waiting oh well it's um you know it's that thing of you you've you still have to understand you know the science behind what you're doing you know and and a simulation or a model can spit

out a number but you know if you don't know what that number means uh not only can it be just wrong but it can be dangerous you know that's that's that element it's it's you can be confident that you've got an answer but is it credible right you do you have an answer that makes sense that that is actually going to make an impact or you know in the the manufacturing World it you know it could be you know you you're actually going to say you would design a product that can't be designed right or wouldn't work um you know and it it's a huge

problem uh and on top of it hmm we're compounding the problem because the Technologies underneath to do the computation are rapidly diverging right it you know it's no longer a CPU world it's a CPU plus GPU world and now there's there's fpgas there's different types of accelerators and there's different you know different things coming online rapidly there's all kinds of different performance libraries and utilities and things that all need to be coming together and it's it's just getting more and more complex and that you know eventually you get out to a user base that's already averse to having to learn a lot of things they

just it turns them off completely right and people move away from using the simulation so it's combating that abstraction of complexity and making it consumable is a key industry element that yeah I know I will upset people uh the HPC crowd has has not really done a good job with with right yeah yeah and beyond that I think um it illustrates so we've talked about the user familiarity but also the code that's out there is often as you say not at all optimized for current generations of computers you know I have single threaded things and users pounding on me to let them run two weeks

in a row which I keep telling them is a bad idea for a lot of reasons but uh so I can't let this go by without mentioning that the NSF cssi proposal for those of you who are in the field uh which is uh computer science software infrastructure uh there are plenty of opportunities for people to write proposals and get involved with making the code better and I would be delighted if if you would that's not new till December so you have plenty of time I don't give a link to that that we could post uh yeah I'll find one yeah thanks that'd be great

I'm gonna go over to Forest in essence you are probably the closest to this from an educational standpoint of the most recently coming out of Education I know that you spent a lot of time around this you're passionate about education and how people come through this field what did you see at going through school and then as an admin from people's desire or wants to learn this stuff so interestingly some of the earliest people I interacted with that were interested in kind of Automation and reproducibility in containers and stuff like that where people within kind of the bioscience I believe specifically the bioinformatics sphere um

I worked a lot with a lab while I was in Boise State that was very interested in modernizing their HPC efforts so they were very interested in things like I said like containerization and next flow and things like that um so at you know a high level there are definitely people you know kind of at a PhD level at you know the pi level that type of thing that are really interested in creating these systems and are you know interested in seeing what can be done in this sphere of computing to improve the operations uh you know for themselves and the people that are working

in their lab um so yeah I I like I said some of the first people I ran into that were really interested in you know taking like uh you know conda and figuring out how to convert their conda environments into containers that they could deploy um we're in biosciences in general um I definitely see a lot of the push towards abstracting a lot of this and I think a lot of that overlap um kind of comes from the fact that much like Ai and this is a you know obviously a domain science that's overlapping with another entire field that people build careers around um and

so you know as difficult as it is for you know people that are dedicatedly in HPC I shouldn't say it's difficult but as inefficient as it would be to have people in HPC trying to learn all these you know domain Sciences to help optimize it's inefficient to have domain scientists trying to learn all of these different things of computing and unfortunately there are side effects to that you know we end up uh you know maybe an example is with you know all the different ways that you know python installs and things like that and packages and things like that are managed there we end up

kind of creating all these different paradigms over time when we kind of leave a lot of this Computing effort to The Domain scientists so it's definitely important that um you know like I said there are people at all levels of the experience some people are looking for reproducibility some people are looking for containers some people are looking for Jupiter notebooks regardless it's important that we make sure that we're doing all we can as dedicated HPC people to apply the knowledge that we have to these domain science problems in a way that like I said optimizes for uh the domain scientists not having to spend tons

of their time learning these specifics of computing which is we would not want to have to learn the specifics of all their fields if that makes sense it does thank you so one of the things I keep hearing that Glenn kind of started out talking about so I'm going to pick on him first but I think it's going to be really interesting to hear John Gary your your take on this also Alan you said it earlier that bioscience has a big data problem so there's a data issue where does that come from um a couple of different places uh real quick though uh to piggyback

of what horse was saying and I believe Alan brought this up at a prior uh prior webinar um the the the rise and Advent of the research software engineer is is is exactly due to what myself and Gary and force just said right there's this gap between you know researchers want to do research and researchers who also know the coding and in that you know or Engineers that know the coding and now we're getting you know more people like bioinformaticians that will just step into that role and say okay biologists tell me your problem and I'll code it for you um and um but you

know you can go look that up at the other prior uh prior webinar but you know they're forming unions now right and they have large online organizations and having meetings whether it's in the UK or the US so um I think that'll continue to continue continue to happen um do you get to the the data um issue so you know now there's a couple of them you know DNA sequencers is is a major one that used to be the biggest uh kind of uh data Creator in the space um you know and an illuminous sequencer if you do run a human genome and aluminum sequencer

now like the raw data that comes out as a terabyte and you'll end up with like after Varian calling maybe it could be like a half a half a terabyte um bam file a VCF file a couple hundred gigabytes so if you want to do a thousand thousand people study you know it gets really large but what's actually because I think be creating more data in the long run you know Gary mentioned the cryo-em studies turn out multiple terabytes per experiment uh in data and the uh there's a lot of Imaging out there that a lot of artificial intelligence has been applied to so whether

it's x-rays or MRIs or CT scans and other things you know we're giving we're giving all that data to train AI on picking out you know uh broken bones or you know Alzheimer's or different kinds types of uh cancer from from biopsies so um there's that's that's the majority of it um I'm sure there's I'm sure there's others that I hope other people will chime in with yeah Gary well you know I think Glenn did a pretty good job I I don't think I have anything to add add to that um oh EHR right EHR records yeah yeah right the hospital records themselves right I

forgot about that one yeah John I gotta believe this is a problem you have to deal with yeah it's a huge problem I mean huge no pun intended um the the shift for me has been from back you know when we were fresh on the heels of the Human Genome Project and in in my head at least I was thinking when you're gonna store a genome you're gonna store a human genome and for all of us there's going to be one human genome stored that's the storage problem and that is so naive and stupid it's ridiculous because your genome changes over time and throughout your

body like you have many genomes roaming about your body right now and so the problem is beyond exponential and how much data there is to store and that's before you take into account the fact that you're not just your genome your your genome plus the Genome of every microbe living on your skin and in your intestine that's you know as an organism the problem just grows and grows and grows now you want to do a longitudinal study you want to collect that genome every month for 10 years and see how things change over time how the microbiome changes over time so the storage problem just

in genomics is enormous and it's only going to get worse and people don't want to just stick this stuff in deep archive and never look at it again and have it just in case they want it online so that the next upgrade to blast comes around I can rerun all my pipelines start to finish on all my old data so you don't just have a data storage problem you have a keep data online in a place where I can get to it and rerun an analysis problem yeah and then that's just genomics throw on top of that the fact that every year the amount of

pixels a CCD can capture goes up and people put them on faster and faster microscopes and want to catch live streams now not just individual images and post-process those with gpus and AI whatever uh yeah this is uh I I don't see us ever actually getting on top of the storage problem that we have that's gonna that's coming down the line and what one thing I didn't mention uh proteomics 2 I think is going to have a lot more data and it's similar to what John was talking about is like now now you want to look at the metabolism going on in your blood in

a person's blood right over a period of time during with and without treatment then you got to compare that to a population it just yeah it just fibonacci's out of control and what what also doesn't help is you know I've been a big proponent of um compression for for genomic data uh so we could store more in a smaller smaller uh you know space but the storage vendors don't want that right the storage vendors keep you know I I would bring up compression that are like shut up go away You're killing my sail kind of thing and uh and I was just like don't you

think the customer would want us you know storm twice as much data uh on on your storage than you're letting them on about and then they would keep buying your storage but like that you know so the incentives there were just completely uh you know at odds to each other so it's um it is a real problem um and the cloud in some ways was meant to deal with this right you could put out this data that that everybody could you know share um but as John said it's just like the amount of it and the types of it is is really just you know

daunting I feel like I thought about it like you said John from The genome perspective but if you're doing just take an individual from the human genome what size are we talking about if you're having to do that for like you said on your skin in your gut what size for just a person are we talking about in that at one point in time that the the size will depend on the sequencing technology used and how much of the original data you want to keep if somebody um way back you know 10 years ago we you may have had people who didn't just want to

keep the sequence data they want to keep the images from the sequencer because the sequencing technology was changing and there may they may be able to reduce Errors By reprocessing the original images and in those cases you're talking about multiple terabytes of data that you would capture from the instrument if you just want the raw faster files those compress really well and you wind up with megabytes to low numbers of gigabytes for an individual person's sequence of data so it just really depends on how much of what the instrument captures that somebody wants to keep to maybe reprocess later or if they just want to

keep the FAFSA if you to get an idea of how big things are you can look at ncbi any of the reference databases because they're just storing the actual sequence and maybe some maybe some quality scores or you know something that goes along with that sequence but for people that have instruments that are collecting data off the instruments a single run can be in the terabyte two terabyte range for an organism like a human and that'll depend on the coverage that they do you know do they do a 50x coverage or 10x coverage or whatever I mean that any multiple of the number of times

they want to to run a particular set of sequence so it's really variable and depends on what the people want to do with the product Downstream yeah so I'd like to I can't resist the ears to mention that my favorite definition of big data is in which to answer your question uh I can't remember who said this but I've quoted it ever since big data is everything we used to throw away yes yes or receipts it's all the video from how long you linger in the Shelf in front of the canned peas and you know so now you uh couple that with the whole idea

that have revolutionized the biological sciences you know we they used to talk about junk DNA now they realize there's no such thing right so it's all information so I think I'm going to make a prediction here I think that we're not far from if we're not already at the point where we can no longer regard these instruments as just raw sources of data we're gonna have to bring intelligence to the instrument and as a particle physicist we got a you know a few decades jump start on you and generating too much data um you know LHC Collision rate is 36 million Crossings per second and

each one has multiple interactions and each event uh can produce uh many gigabytes of data so um there's no way that we let all that data just come out of the machine right out of the instruments that are surrounding the collisions we have for decades put many levels of processing that we call the trigger uh my favorite definition of the trigger was from a colleague at University of Wisconsin that said trigger is designed to throw things away so you know it's couples with the previous definition right so you only select out of those you know billions of of bytes of data that can happen every

second billions of gigabytes of data that can happen every second we only select a small tiny fraction of that to write to storage and to distribute the worldwide computers I think still the largest single uh computer infrastructure is the LHC Computing grid which I helped design so I'm kind of proud of it so but I think the prediction here is is is not a competition for data it's that you're going to have to bring intelligence to the instrument so you're gonna have to make those decisions those image analyzes and the the design of your experiment before you come to the instrument uh when you run

a particle physics experiment you design your trigger before the collisions happen you decide what you're going to keep before you you actually turn the thing on and I think that we're going to get to the point with these instruments that you can't keep every single frame of every single image scan that you do you can't keep it all the output you're gonna have to design the data reduction to happen there and still you will have a huge amount of data so that's an interesting point Alan and it kind of leads into the next topic here if we talk about HPC and performance and why it

matters in bioscience to what you just said the more performant that we can get does it lead a researcher to want to collect more and analyze more or it's kind of going the opposite of what you said what we need to do but are we going the other way so let's talk about HPC performance and why does it matter in bioscience I was like anyone who wants to talk Jonathan you haven't talked in a while yeah I I something Alan said sparked a question in me so so I I love I love um I love that quote that you that you gave that the big

data is everything we used to throw away and I think it's no coincidence that big data has really at least from my perspective I'm not a bioscientist uh but my my experience has been and my observation has been that big data has come up in response to the Advent of machine learning and that it it the the incentivizing value function of big data has been that when we write learning algorithms they find things in the data that we didn't understand were there in the first place and that I read an article recently I forget where it was but the the idea that we keep making

the mistake of trying to teach the computers what we already know instead of allowing them to learn things from first principles and just by pointing them at a bunch of data uh and that every time we do that they had a discovering things we didn't understand were there in the first place and we slowed that process down by trying to encode our own value functions or our own understanding of of the the area that we're researching ahead of time and I'd be curious like to me hey Zane your question was about performance but to me the performance here is less an issue than the storage

problem which is is still part of the same thing in in my mind like the capability of the resource how much can we store how fast can you get through it so are we going to have a dolly and stable diffusion making up genomes now yeah I don't know I'm asking like if if it is true that we keep learning new learning things that we didn't know before in the data that we thought we could throw away before like how do we make that decision how do we possibly like that that's what's incentivizing keeping everything because we keep finding stuff in it that we almost

threw away right I don't know what to do about that okay so so there's two lines of discussion I want to bring up one is one is really fun because it's it's it's biologists versus physicists um Alan brings up a really good point and my first thought was uh you know on the instruments well at least I'll just be talking about like Illumina sequencers right they you can you can you could say hey luminous sequences genome and they'll send you back the the file with the mutations in it right just the VCF file um but in order to accept that BCF file you have to

agree with how they did their analysis right and the problem Illumina runs into all the time is like the rest of the research Community says no your algorithm doesn't find the indels and and the deletions properly or it doesn't it doesn't do the copy number variation properly so I have to redo this myself anyway so there is definitely a data reduction process happening I just think with our algorithms and our knowledge of the genome now to Jonathan's point is like we don't have it systematized enough or we can just say this is definitely not you know a mutation or this is definitely the right copy

number of variant variation um but but hopefully we'll get there right with it with enough with enough um enough data and enough runs um so biology is a system but we just don't understand it as well as as a lot of the physics systems out there so um and the second point was to get to the performance bit um you know again Illumina has uh bought this fpga company at Echo genome that I used to work with and they uh they speed up that analysis uh several folds right it used to take let's say uh well it could take it you know depending on size

your cluster like on the on the cluster I used to run on previous employer would take 12 hours to run a variant calling um analysis just on CPUs um but edico genome on their fpga could do it in under half an hour right same 40x coverage so really really fast so Illumina is taking these cards and shoving it into their sequencers to say hey we can get you that you know that result like that now problem with that is is the fpga's code is locked you can't see it so so you so do you run it against your CPU code anyway to see what what's

going on um so we're really in this battle of of like you said of trying to you know time to treatment right like if I can run it on an fpga that uh you know six seven times faster I can get to a drug for patient you know it in inside of a day like any other kind of like clinical lab test which is what we want um but what what one group will throw away you know data is somebody else's you know someone else's garbage is someone else's what is it dinner I don't know um and they'll say you're throwing away the wrong things

we need to look at this so I think it's I think it's a really big issue and it's being tackled I think it's just not um we're just not we just don't have an understanding as well as Jonathan brought up about like what we can all agree upon you know to keep or to throw away and and make and maybe maybe that'll will never agree upon it hopefully that that Gap shrinks right but um it's a really big a big problem um as you know the data gets larger and and we're trying to get to treatments faster from uh from the perspective of uh the

responsibility should be on the person design the experiment to throw away data as part of the experiment there is actually some relief coming from that somebody was telling me just in the last couple of weeks about one of these uh single molecule long read sequencing Technologies that's up and coming uh they'll have the ability that when you slice your sequence to put it into the sequencer to tag it so that when that single molecule reader picks up that sequence it'll look for your tag and your tag will only be attached to the strands you care about and it'll immediately throw away the things that you

don't care about so on the chemistry side itself in the instrument it'll be able to filter out and just look at strands that you have tagged that you want to sequence yeah so if we spun real quickly to the uh the data volume question because I think I was actually trying to make a point um that's serious here which is the um yes the design of how you keep data is uh is crucial and when you design triggers in particle physics you spend a lot of your time designing cross checks and so forth but my my statement my I'll stand by it is I think

it's inevitable you're going to have to learn how to do this it's not going to be the case your date your instruments are getting better and they will simply generate too much data so I think it's something like three and a half or four terabytes uh a second that LHC can can process uh so um uh the lacit systems can process as intake so you know you have to get it down to that level and the statement is that the instruments are improving fast enough as as was pointed out that um this will have become an inevitability I give an anecdote an anecdote that will

refute that uh when I was at the at the broad Institute a very very long time ago we were running out of storage and we pulled in the sequencing group and said you need to hire someone to go through your your six petabytes of data and figure out what we can throw away because we're running out of storage or we need more money for storage and without hesitation they said here's a million dollars go buy more storage yeah yeah so yes but I still think the instrument's a good grow fast too yeah in storage well well the the collider has I think a really is

a really good example of that the the data is coming so fast that you you can't capture it all or are you going to lose the valuable stuff but that you know we probably should have scheduled about four days for this webinar because no kidding there's so many people the part about the LHC I'm really uh uh envious of is is this one project right and everything kind of trickles down out of that if we only had one Sequencing Center like the broad in the U.S it might look better like that and the broad still has dozens of petabytes of data out in Google Cloud

but um I and I think that's part of the issue right there's these multiple large data centers and multiple unit many universities have their own sequencing centers so um and that's good because we're coming up with you know better algorithms for to do a lot of these things but you know you're right I think the instruments themselves will just you know like uh John was saying neither you're gonna have to go back just to targeted regions of genomes that you're interested in or uh or some other kind of triage of data um hopefully um or or be able to just process more which is really

where the the gpus and the fpgas have really have really helped us in the in the past but I think I don't know that I don't think that we can rely on you know more GPU cores or or new silicon to be able to get us out of this problem by itself yeah that'll that'll work except for people who are you know they want to just put a sample in uh you know take your dirt wash all the DNA out of it put it in there sequence every Strand and then blast that to see what was in the original sample agreed yeah there's always going

to be somebody who wants every strand of DNA out of a sample sequence so that they can see what was in the sample yeah and we and we're doing that with uh with uh with tech with people coming who are sick and we don't know why they're sick right we just everything gets everything gets swabbed and sequenced right and now if you want to do that for a million people in the next pandemic you know you're not throwing anything away either there either so yeah it's a it's it's it cuts both ways Gary I didn't get something you want to add you know I I

I'm just thinking about what Alan was saying about processing the data instrument I mean is it definitely an HPC problem I mean at the lab right now where uh there's a project to do a gamma ray detector uh that's going to be run in conjunction or the radio radioactive isotope accelerator and uh but the detector itself is going to generate so much data that we're actually using an HPC cluster for so this is an example of like processing we're using an HPC cluster as the first level trigger of the data coming off the detector to calculate the path through the detector and what hit segments

it hits um and just sizing that and and figuring out like how many gpus you need that certainly a HPC problem and one one advantage you know a design decision that they did of using a cluster instead of fpgas is that the the thinking is that we could use different containers depending on the experiment or what you know what we're going to be running through the detector so it gives us some amount of flexibility so that this way you're just not locked into one way of throwing away data thank you I think we have one question and I've got another question for Glenn and Gary

so what do you think the biggest challenge in genomics medicine what is the biggest challenge you know genomics medicine so outside of the size and the growing size of data actually I think it's who owns it when it comes to the when it comes to the to the genomic data if you get your your genome sequenced uh in the past it's without a doubt it's been it's been resold or or used for other purposes 23andMe uh you know all these other all these places that we'll sequence data typically will also kind of sell it to Pharma companies to you know because they want to create

they want to create find people to do trials on and stuff like that people have found that not to be really palatable lately and they want to own their own data there's a problem there though um because they've got hospitals and clinicians and Physicians that say you know have legal liability things like that so it my prior employer we were there were projects where we were recruiting people like in Alzheimer's studies therefore we're going to do their sequencing there we're going to be able to to own that and only be able to opt in uh if they wanted their data shared to other companies that

may you know maybe looking to create cohorts for example um so I think that I actually think it's kind of like that ownership um and privacy issues that are going to be going forward because how do you protect some people's data and not others right and you know and you get to varying levels of what you can share with other people or other other companies and and so I was also working with a company that was trying to like you know be able to encrypt this data and send it around and only give it only give people access to certain parts of a genome for

these reasons and so for me that you know if we if we don't want kind of genome data to go the way of social media I think it needs to we need to get our hands on that and I don't I don't see any leadership either but I'm not really kind of you know tied into tied into DC about what's going on with these things but different folks are trying to different groups are trying to do that um it just seems to be kind of parochial at this point this is the Privacy issue the Privacy issue uh is one that kind of I find kind

of fascinating and funny uh there is whenever somebody talks about anonymizing sequence data it blows my mind because there's nothing more uniquely identifying than your sequence it literally cannot be anonymized right because it points to you and we're shedding it as we walk through the environment we are shedding that unique identifier everywhere we go so the very idea that we try to keep secrets data private I I find utterly ridiculous because it's just a matter of improvements and sequencing technology until I can sequence You by picking up your Starbucks cup that you just threw away and know your sequence and follow you around town right

you know with my phone probably mentioned Hawaiians used to have a Prohibition against Gathering the hair clippings or um any other material from the from the royalty uh maybe they were onto some something yeah I mean this this is a problem it's too late to address it as a privacy problem it needs to be addressed as a what's legal to do with your data problem not how to keep your data safe because I mean we're all on the internet your data is not safe there is no such thing as safe data it does not exist yeah it's very true I think kind of going back

Glenn we've talked about instrumentation instruments in a bio environment what are some of those instruments where are we going with them I mean it's for me who's not in that field I mean I can think a microscope you mentioned that earlier but what other things out there Gary pointed back to what I'm not going to try to repeat what he said but a very complicated using radioactive eye steps what are other tools out there that we're using um so one that's creating a lot of data is the in proteomics space is the mess spectrometer um which used to only well so and I think the

reason why we haven't seen a lot of that data on the HPC side is is because the the programs used to analyze that data are all shut down proprietary by Thermo Fisher and they you got to run them on a PC but uh uh so so Mass Spec is one um the sequencers are the sequencing Technologies is uh moving towards I think um you know long read sequences um I think the thing that impresses me most most is something like that John brought up is that you know I can get a DNA I can spend a thousand dollars now and get a DNA sequencing kit

uh sent to my house right and it's the same one that they put up on the space station um uh and ran you know ran genomes genomes out up there so I think they're just gonna get you know smaller and more convenient to use and you know if I can sequence a genome for a thousand dollars you know in 10 years can anyone do it for a hundred dollars and what do we do when everyone's sequencing everything in their backyard right so um well I think icloud's going to look look a lot different but um yeah I haven't really thought about other sequencers I really

think about data capture of you know you're swallowing yourself you're swallowing the ill now it sends out signals to either patches on your arm or to Bluetooth and people are collecting that data as well about your heartbeat your blood pressure whatever it's monitoring um that seems to be more not only invasive but just more data more time points uh and if anyone's looked at Fitbit data that gets that gets transmitted it is just garbage I mean consistency wise I mean it's just so noisy so um I think there's those are the kind of problems I think we're we're going to see more of um yeah

and and I would I would add you know I I joked uh you know asking when am I gonna get to meet my digital twin right and so digital twin is a a model representing a physical object and it's you know it's very big in manufacturing right now but you can imagine where we're headed is everybody having a digital one or more digital twins that exist that you use a digital twin to to simulate you know things happening right you know and then again in manufacturing like an airplane engine they they can do digital twins of an airplane engine and you start to see when

does the digital twin fail why does it fail and you actually fix the engine a problem the engine before it ever fails right well this is the problem that Alan's talking about when you get to digital twins of people the micro elements they may be sequencing themselves you know all the time that that that pill you swallow that goes through and reports for five hours well now is 24 7 and that's data just flowing and you've got to make those decisions you know what data do you keep you know what what just gets thrown away but it's it's exciting and scary to think about you

know for all of those reasons you know it's it's data about you it's it's everything that identifies who you are what you are and then what you do with that data you know if if we saw everything about ourselves and and you know how we might respond or even die you know what does that do to our psyche yeah it's amazing to think about but it is I'd like to to challenge the audience to uh to write the words digital twins of people down on a sheet of paper and uh sit down with their favorite screen script writing program and generate our next science fiction

content for us before it happens yeah if I've got a digital twin kind of the Brock's point I want them to actually have a seat on an airplane that fits me I want to be comfortable and remember a lot of that a lot of data that's getting transmitted we talk about pills that get swallow or your Fitbit or whatever that's going you know Bluetooth over home networks right yeah and even if it was inside a hospital I mean those networks and everybody's home networks not great security-wise but it's just not great so um so that's that's a concern I mean you think about it like

you're reading off stuff on Alexa right to to into your doctor's office or or to iCloud where you're where you're Dr CRM or epic has access to it that is terrifying from a from a hacker's perspective even even if privacy actually still existed which it doesn't people would happily sign it away in the EULA to get access to the next video in The Tick Tock stream yes they would it just it just doesn't exist anymore there's no such thing as privacy and it's it's crazy to see people going on and on about how we need to worry about privacy given the state of the world

yeah that's a good point well guys I really appreciate it we are up on time and I feel like we could just keep going for a couple of hours and keep drilling into this so hopefully come back and we'll pick another topic but we'll continue this discussion I really appreciate it everyone out there watching appreciate your time thank you for joining us go ahead and like And subscribe and we will see you next week thanks guys thanks