Transcript: sfspark.org: Alexy Khrabrov interviews John St. John

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