sfspark.org: Alexy Khrabrov interviews John St. John

hello everybody I'm Alexa crabber for hello everybody I'm Alexa crabber for the organizer of SF spark and France and the organizer of SF spark and France and here we are on location a chartboost for here we are on location a chartboost for a very special meet up on open genomics a very special meet up on open genomics it's called scala and spark for good it's called scala and spark for good open genomics mobilizing open source open genomics mobilizing open source developers to work in cancer research developers to work in cancer research together and here was who have johnson together and here was who have johnson john from driver genomics he's one of john from driver genomics he's one of the folks who actually introduce our the folks who actually introduce our community to the field they are heavy community to the field they are heavy users of Scarlett's Park and I let John users of Scarlett's Park and I let John introduced himself and talk a little bit introduced himself and talk a little bit more about it thank all I see ya so I more about it thank all I see ya so I work at a company called driver group I work at a company called driver group I was one of the co-founders we are was one of the co-founders we are currently using spark and in Scala to to currently using spark and in Scala to to start looking at mutations and in in start looking at mutations and in in giant data cancer samples and so giant data cancer samples and so basically what we do is we collect we basically what we do is we collect we collect samples from a tumor and a collect samples from a tumor and a normal from a cancer patient and and we normal from a cancer patient and and we do a comparative analysis of those two do a comparative analysis of those two samples and with that we produce a free samples and with that we produce a free report that recommends basically which report that recommends basically which kinds of drugs those patients should go kinds of drugs those patients should go on and what our company is doing is just on and what our company is doing is just giving these reports away for free in giving these reports away for free in exchange for data and we're building up exchange for data and we're building up this data and working on identifying this data and working on identifying drugs that will hopefully help these drugs that will hopefully help these patients in in ways that they're not patients in in ways that they're not currently getting help so we're starting currently getting help so we're starting off with the with stage 4 lung cancer off with the with stage 4 lung cancer and we plan and moving into other tumor and we plan and moving into other tumor types afterwards so what made you settle types afterwards so what made you settle on skull and spark as your developments on skull and spark as your developments tech yeah so you know a big part of it tech yeah so you know a big part of it was was the community just seeing a lot was was the community just seeing a lot of a lot of interests in it and a lot of of a lot of interests in it and a lot of a lot of development it seemed like seem a lot of development it seemed like seem like there's a lot of building momentum like there's a lot of building momentum another aspect was the type safety so another aspect was the type safety so specifically in our in our report specifically in our in our report generation it's really important that we generation it's really important that we don't make any mistakes when it comes don't make any mistakes when it comes to you know the different kinds of to you know the different kinds of identifiers zwi have we definitely don't identifiers zwi have we definitely don't want to you know mix up the the case want to you know mix up the the case identifiers with the patient identifiers identifiers with the patient identifiers and accidentally give the wrong report and accidentally give the wrong report to the wrong person so having that kind to the wrong person so having that kind of you know built-in type safety all the of you know built-in type safety all the way down really gave us a lot more way down really gave us a lot more confidence in what we were producing confidence in what we were producing this is interesting that reminds me know this is interesting that reminds me know so kaiser Health System right before the so kaiser Health System right before the dispensing a drug then to scan your dispensing a drug then to scan your wrist bands this is almost implementing wrist bands this is almost implementing this in software like using the type this in software like using the type system to keep track that you're not system to keep track that you're not mixing our patients exactly days I wish mixing our patients exactly days I wish which is this isn't a kind of an which is this isn't a kind of an interesting angle about the type safety interesting angle about the type safety which is which is I think it's a great which is which is I think it's a great application of type safety so but above application of type safety so but above all the data size and speed why don't all the data size and speed why don't you do it all in Perl right like there you do it all in Perl right like there is there are books in our offer by is there are books in our offer by scientist and Python what you know scientist and Python what you know technologically is different about technologically is different about scotland's parkash makes it easier for scotland's parkash makes it easier for you guys to use them mm-hmm what I you guys to use them mm-hmm what I really like so let's say we're going to really like so let's say we're going to do this all in Perl it would be an do this all in Perl it would be an extremely manual process you know you'd extremely manual process you know you'd be defining exactly where you split up be defining exactly where you split up the code exactly how you paralyze how the code exactly how you paralyze how you move the data back and forth you you move the data back and forth you basically be doing all that really basically be doing all that really low-level work yourself and on top of low-level work yourself and on top of that pearls kind of slow to be honest so that pearls kind of slow to be honest so the fact that Scala itself was a fairly the fact that Scala itself was a fairly fast compiled language and it would do fast compiled language and it would do you know with spark it would do a lot of you know with spark it would do a lot of this parallelization and distributive this parallelization and distributive processing for you it was incredibly processing for you it was incredibly incredibly attractive so yeah so spark incredibly attractive so yeah so spark is basically an easy way to paralyze is basically an easy way to paralyze computation mm-hmm exactly right so can computation mm-hmm exactly right so can you tell us a little bit about yourself you tell us a little bit about yourself and how did you basically come to a and how did you basically come to a local found driver what are you local found driver what are you primarily biologists a computer primarily biologists a computer scientist how do you you know see the scientist how do you you know see the fusion of these two tracks well i would fusion of these two tracks well i would say in academia this fusion is starting say in academia this fusion is starting to become pretty mainstream so there are to become pretty mainstream so there are a lot of programs out there that that a lot of programs out there that that merge the two about fifty percent of the merge the two about fifty percent of the time you're doing biology fifty percent time you're doing biology fifty percent of the time you're doing computer of the time you're doing computer science actually that's what i did with science actually that's what i did with my undergrad my undergrad I majored in computer science and I majored in computer science and minored in biology uh-huh I discovered minored in biology uh-huh I discovered that you know solving these kind of that you know solving these kind of biological problems using computers was biological problems using computers was just really really interesting to me you just really really interesting to me you know i think we had this nice project know i think we had this nice project where I was looking at HIV data at where I was looking at HIV data at mutations in HIV patients and I was able mutations in HIV patients and I was able to quickly find that certain mutations to quickly find that certain mutations resulted in in patients having lower cd4 resulted in in patients having lower cd4 t-cell counts and that was like just a t-cell counts and that was like just a really interesting thing that you could really interesting thing that you could find just by looking at this data that find just by looking at this data that wasn't necessarily obvious and I'm not wasn't necessarily obvious and I'm not even sure if it was published at the even sure if it was published at the time but you know it was it was just a time but you know it was it was just a really fun kind of intro to what you really fun kind of intro to what you could do with with computers and could do with with computers and biological data so it went on from there biological data so it went on from there and I eventually got involved in a and I eventually got involved in a slightly different thing different kind slightly different thing different kind of biological data i was looking at of biological data i was looking at alligators and crocodiles and gharial alligators and crocodiles and gharial xand putting together their genomes for xand putting together their genomes for the first time after you have a first the first time after you have a first genome it's really easy to do a lot of genome it's really easy to do a lot of different analyses but putting together different analyses but putting together that first genome is really that first genome is really computationally intensive and very computationally intensive and very difficult so I was doing that for these difficult so I was doing that for these animals to help other scientific animals to help other scientific researchers with their downstream data researchers with their downstream data analyses and actually that is how I got analyses and actually that is how I got involved in driver uh-huh yeah you say involved in driver uh-huh yeah you say so you measure this locales can explain so you measure this locales can explain through our audience cuando biologists through our audience cuando biologists what the low counts meant in the context what the low counts meant in the context sorry say that again eventually low sorry say that again eventually low counts and so one of your order problems counts and so one of your order problems you found Oh cd4 t-cell counts ah yes so you found Oh cd4 t-cell counts ah yes so basically cd4 t-cells are a kind of basically cd4 t-cells are a kind of immune cell that your body has and one immune cell that your body has and one of the effects of HIV is that these of the effects of HIV is that these immune cells are depleted their killed immune cells are depleted their killed off by the virus so what you can do is off by the virus so what you can do is basically it's it's you know it's one of basically it's it's you know it's one of the things that makes you sick when you the things that makes you sick when you have AIDS so you know you measure these have AIDS so you know you measure these levels and what you want to do is try to levels and what you want to do is try to keep these levels high just so basically keep these levels high just so basically by by correlating the the data basically by by correlating the the data basically the the data of the virus and what the the the data of the virus and what the virus virus sequences with the outcome you know how sequences with the outcome you know how low the cd4 t-cell counts were in a in a low the cd4 t-cell counts were in a in a patient mm-hmm you can you can basically patient mm-hmm you can you can basically what i was hoping to find were were what i was hoping to find were were these correlations basically identifying these correlations basically identifying mutations that would possibly result in mutations that would possibly result in more aggressive forms of the virus more aggressive forms of the virus mm-hmm right so more aggressive kinds of mm-hmm right so more aggressive kinds of HIV potentially interesting or maybe HIV potentially interesting or maybe more resistant to drugs so this is an more resistant to drugs so this is an example basically the data science example basically the data science directly leading to yeah to discovers directly leading to yeah to discovers exactly interesting addressing and you exactly interesting addressing and you know that that was a toy project it know that that was a toy project it wasn't anything that you know turned wasn't anything that you know turned into anything useful but still it was into anything useful but still it was just really interesting seeing that kind just really interesting seeing that kind of data and just exploring it myself so of data and just exploring it myself so I can you like doing on this there is I can you like doing on this there is this term personalized medicine right this term personalized medicine right and you're in the frontier of this yeah and you're in the frontier of this yeah so what you're building essentially an so what you're building essentially an example so can you explain and summarize example so can you explain and summarize for folks new to the area what is the for folks new to the area what is the personalized medicine promise and you personalized medicine promise and you know how you specifically going to know how you specifically going to achieve this in your space right so you achieve this in your space right so you know we're focusing on late-stage lung know we're focusing on late-stage lung cancer right now and I'll just go into cancer right now and I'll just go into this with a with a case so one thing this with a with a case so one thing that came up a while ago was was that that came up a while ago was was that there there about the about seven there there about the about seven percent of lung cancer patients have a percent of lung cancer patients have a mutation in a particular gene this gene mutation in a particular gene this gene is called egfr so of those patients you is called egfr so of those patients you know you could just give them chemo and know you could just give them chemo and radiation in that case they'd have the radiation in that case they'd have the same outcome as any other patient any same outcome as any other patient any other lung cancer patient which is you other lung cancer patient which is you know not very good by knowing that they know not very good by knowing that they have this particular mutation and by have this particular mutation and by knowing that that is the mutation that's knowing that that is the mutation that's driving their lung cancer that driving their lung cancer that particular mutation is causing their particular mutation is causing their lung cancer to happen there are actually lung cancer to happen there are actually drugs out there that match up with that drugs out there that match up with that particular mutation so these drugs will particular mutation so these drugs will actually target that gene and turn it actually target that gene and turn it off so now the tumor can't grow anymore off so now the tumor can't grow anymore so when we talk about targeted targeted so when we talk about targeted targeted therapy what we're talking about is a therapy what we're talking about is a therapy that is very specific to an therapy that is very specific to an individual patient right this is the individual patient right this is the personalized aspect of this kind of personalized aspect of this kind of therapy so for those patients therapy so for those patients now that have this drug that's matched now that have this drug that's matched up with their specific kind of tumor up with their specific kind of tumor they they do on average about six months they they do on average about six months they have about a six month longer life they have about a six month longer life expectancy than patients that just do expectancy than patients that just do standard chemo and radiation and you standard chemo and radiation and you know although that I mean that's know although that I mean that's actually quite a long time for these actually quite a long time for these patients six to eight months extra for patients six to eight months extra for them to potentially you know find them to potentially you know find another therapeutic option or you know another therapeutic option or you know just finish up what they need to finish just finish up what they need to finish up in their life I mean it's it's huge up in their life I mean it's it's huge for them even though it's you know start for them even though it's you know start right a minute yeah given the type right a minute yeah given the type frames yeah usually that's that's so so frames yeah usually that's that's so so basically this is a direct result of basically this is a direct result of doing proper match of individual keys doing proper match of individual keys yeah and available already available yeah and available already available minutes so this a quiz today yes right minutes so this a quiz today yes right so but going to the future ah can we so but going to the future ah can we kind of see a world where have much more kind of see a world where have much more drugs and kind of match them drugs and kind of match them individually what would it take you know individually what would it take you know together like doing it just is it just together like doing it just is it just the commutation problem or is it also a the commutation problem or is it also a government problem like how the pharma government problem like how the pharma companies fit in there and what is your companies fit in there and what is your plan to a nickel this yeah yeah I think plan to a nickel this yeah yeah I think there's a really big collaboration there's a really big collaboration problem right now where you know every problem right now where you know every every doctor and every lab wants to have every doctor and every lab wants to have their own data and wants to be the hero their own data and wants to be the hero and discover something so you have are a and discover something so you have are a lot of different siloed data sets that lot of different siloed data sets that are very specific to different centers are very specific to different centers and these silo data sets they don't get and these silo data sets they don't get shared you don't have a lot of people shared you don't have a lot of people collaborating on discoveries there are a collaborating on discoveries there are a couple public efforts that are you know couple public efforts that are you know working on getting past this but they're working on getting past this but they're they're pretty minimal they're not you they're pretty minimal they're not you know you don't get hundreds of thousands know you don't get hundreds of thousands of patient samples you get maybe you of patient samples you get maybe you know 1,000 or 2,000 in a particular know 1,000 or 2,000 in a particular tumor type and that's those are the tumor type and that's those are the bigger projects that are out there so I bigger projects that are out there so I think that is partially holding the think that is partially holding the field back so I think that you know field back so I think that you know something that really that opened up the something that really that opened up the data and also you know open data and also you know open collaborative potentially even collaborative potentially even competitive analysis analysis competitive analysis analysis consortiums you know like different consortiums you know like different kinds of competition else I'll kinds of competition else I'll competitions competitions I think those kind of things would be I think those kind of things would be huge right because the alternative is huge right because the alternative is you publish your own paper and you run you publish your own paper and you run your own competition and there is so your own competition and there is so much room for either you know direct or much room for either you know direct or indirect dishonesty or just like you indirect dishonesty or just like you know not trying is hard to make the know not trying is hard to make the other people's methods work and you know other people's methods work and you know it doesn't really push the field forward it doesn't really push the field forward not to mention duplication of work right not to mention duplication of work right so this is actually very nice the kind so this is actually very nice the kind of situation to our meetup topics so we of situation to our meetup topics so we we had this informal group start as a we had this informal group start as a text by the bay where we have folks from text by the bay where we have folks from two universities and in three startups two universities and in three startups right I think most of them are here right I think most of them are here today and we have a wider group of today and we have a wider group of developers and the response we should developers and the response we should get you know presenting them with this get you know presenting them with this is always very enthusiastic so basically is always very enthusiastic so basically all the people say how can I help of all the people say how can I help of nominees sake how can we help for nominees sake how can we help for instance nitrile my company has a instance nitrile my company has a program called natural gifts where we program called natural gifts where we give developers and the employees in give developers and the employees in general five days a year to volunteer general five days a year to volunteer want projects of their choice and and want projects of their choice and and also in as a company we can put the also in as a company we can put the resources behind it's right and I think resources behind it's right and I think a lot of companies in the area and a lot of companies in the area and beyond have this kind of programs right beyond have this kind of programs right so so potential there is a lot of so so potential there is a lot of resources and individual time you know resources and individual time you know in your spare time you can choose to in your spare time you can choose to build a pet store under bed rails or you build a pet store under bed rails or you can help with Adam in Scotland spark can help with Adam in Scotland spark right and we hope that we can enable you right and we hope that we can enable you but you know Adam and spark is not pet but you know Adam and spark is not pet store right like that everybody in store right like that everybody in sounds you know how you can buy cats and sounds you know how you can buy cats and dogs not everybody understands how are dogs not everybody understands how are you you are identified mutations so what you you are identified mutations so what do we need to do collectively how can I do we need to do collectively how can I go about providing these developers with go about providing these developers with resources and let's say what if let's resources and let's say what if let's say you have a you know a thousand say you have a you know a thousand developers in your disposal which i developers in your disposal which i think you will give them the trajectory think you will give them the trajectory and basically but for now you can have and basically but for now you can have them through you know spare time and can them through you know spare time and can if you will have partial time of if you will have partial time of thousand developers nights and weekends thousand developers nights and weekends and and an open source set up what how and and an open source set up what how what's the most effective way to utilize what's the most effective way to utilize you know these thousand developers how you know these thousand developers how do you structure their work how would do you structure their work how would you structure the community hub so we you structure the community hub so we can can kind of separate tasks publish kind of separate tasks publish interesting things what are the good interesting things what are the good starting points will this community starting points will this community resource okay so I think what would be resource okay so I think what would be amazing is if we could get everyone amazing is if we could get everyone doing Kegel style competitions releasing doing Kegel style competitions releasing their code and putting their code onto their code and putting their code onto platforms that are really easy to you platforms that are really easy to you know integrate into large like you know know integrate into large like you know enterprise style environments mm-hmm so enterprise style environments mm-hmm so if they could do their code and in if they could do their code and in Scotland spark the first time that would Scotland spark the first time that would be excellent even if they just implement be excellent even if they just implement their algorithms and other languages and their algorithms and other languages and then other developers can come and re then other developers can come and re implement them in scholar spark that implement them in scholar spark that would also be excellent but I think would also be excellent but I think getting more people participating in in getting more people participating in in the competitions that exist and also you the competitions that exist and also you know generating new competitions which i know generating new competitions which i think is going to be you know there's think is going to be you know there's going to be a lot of you're going to going to be a lot of you're going to have to get a lot of like you know have to get a lot of like you know government approval for this you're government approval for this you're going to have to get patients that agree going to have to get patients that agree to release their data and you know open to release their data and you know open ways there's a lot of work to be done ways there's a lot of work to be done but getting more people trying out their but getting more people trying out their ideas on these data sets and trying to ideas on these data sets and trying to make discoveries and trying to you know make discoveries and trying to you know just also just getting the plumbing to just also just getting the plumbing to work and you know like you said work and you know like you said contributing to a demand no avocado and contributing to a demand no avocado and these other tools just getting that data these other tools just getting that data movement stuff you know before even the movement stuff you know before even the data science aspect of it I think those data science aspect of it I think those are all important and this is the area are all important and this is the area we shall lot of that engineers can know we shall lot of that engineers can know very well all right so I think that can very well all right so I think that can be an entry point for a lot of guys here be an entry point for a lot of guys here how is the top the data pipelines right how is the top the data pipelines right so this is great so we're looking so this is great so we're looking forward to your token well configure for forward to your token well configure for the developers to collaborate with Frank the developers to collaborate with Frank and John and get this field moving and John and get this field moving forward we have the new committee forward we have the new committee resource called open genomics at i/o resource called open genomics at i/o will published all the links there so if will published all the links there so if you want to participate join our mailing you want to participate join our mailing list follow the discussions and list follow the discussions and hopefully we'll get this whole committee hopefully we'll get this whole committee going