Contributed Talks 4
Anthony Mammoliti (Princess Margaret Cancer Centre)
Shraddha Pai, PhD (Donnelly Centre for Cellular and Biomolecular Research, University of Toronto) Postdoctoral Fellow
Gabriel Odom (Florida International University) Assistant Professor
Ruth Schmidt (Plotly & Institut national de la recherche scientifique (INRS)) Microbial Ecologist & Data Scientist
3:00 PM - 3:55 PM EDT on Thursday, 30 July
ORCESTRA: a platform for orchestrating and sharing high-throughput pharmacogenomic analyses
netDx: Building interpretable patient classifiers by multi-omic data integration and patient similarity networks
A global test for integrative pathway analysis of multi-omics data
Visualization of multi-omics data in microbiome research
Computational methods to build biologically-grounded, interpretable predictors for clinical decision-making. Particular interest in brain-related disorders.Перейти в профиль
I am a microbial ecologist turned data scientist who makes big data more accessible and informative. Whether it is through creating interactive visualizations, dynamic reports, or user interfaces it is my mission to turn data into actionable insights. I am curious, creative, and collaborative and enjoy working in a multi-disciplinary environment.Перейти в профиль
Where can I read the fourth? Contributed talk session of five conductor. We have four wonderful speakers, given 10 minutes, oxidase followed by a roughly 15 minutes. Please submit your questions filed, the possible whole feature. And make sure, you know, with speaker, your question, our first Speaker today is Anthony's mom, Avicii. I will start that video. And dr. Benjamin a teen blog, a Princess Margaret cancer centre and today alongside with manure Rue who the software developer in the lab. I will be discussing our classroom call
Orchestra, which aims to achieve research reproducibility in pharmacogenomics. First off, I would like to acknowledge that there are many current challenges today in the field. There are large amounts of data that is being generated and therefore, the analysis pipelines, and interpretation of the data is becoming increasingly complex. In addition, there are a plethora of new competition of pipelines that are being released two processes, influx of data, which is often very difficult to keep up with Airport is being said of your pacing and reproducibility crisis. There are
many notable pharmacogenomic datasets over the past few years that have been released. And this includes GTS future pv2 nccoe from The brode Institute Gray from the Joe W. Gray live in Oregon from The Institute for molecular medicine in Finland, from the university health network in Toronto and ggcse 1 and 2 from the Sanger Institute. Essentially, our goal is to utilize panel a cancer cell mice, in a large from academic data sets, which include drug sensitivity, screening and molecular profiling.
4 mile marker, discovery. Therefore, the main challenges. How can we process and share this complex data using a plethora of pipelines in a way that will achieve reproducibility and data provenance? It's all this issue. We developed Orchestra which is a cloud-based platform that allows for the transparent, reproducible and flexible analysis, and sharing, and pharmacogenomics data to Facebook, genomic and pharmacological profile to cancer samples. Through automated inversion processing pipelines using a
school called Pachyderm data object in Reservoir mocha side or be set for future analysis. Using the formal gx5, conductor package, As a reference. Farmers that also known as a tea. Set is a multi-layered data object which includes selling drug Creations drug sensitivity and look at their profile data which all can be used for Bob marker, Discovery and Drug repurposing. At the heart of Orchestra by Pachyderm which is an open-source will that allow for the deployment of
former clients to producing Pieces by storing data in repositories is kubernetes and a cloud or on-premise cluster environments. Words that have Adidas at and peace at can easily be tracked identifiers that are assigned to them by the tool, it was ever changing a repository such as to the modification and this will automatically trigger pipeline employment and therefore we will get a new version of a piece set. Ultimately Orchestra host 7th grade, our
customizable, molecular and pharmacological data on a platform and this gives rise to the exploration of 1676 unique, 928, drug compounds and 1.5 million drug sensitivity experiments. A prophet Oliver molecular data on our on Prime HPC using snake make inversions Oliver data and pipelines every file that exists in one of the deal repositories with inner pocket and plaster is Virgin with a unique ID and is represented by the red circle in the diagram. Put in her pocket and Foster are the process molecular data selling annotations.
Users can customize the content of their former host at which includes molecular data, that is processed with different tools, such as hell. So, and salmon for RAC, theater in different versions of do, responsive for a given data set. However, for each user selection Pachyderm, will combine the various versions pipelines together to form of versions Pista. However, just say that there is an update to the molecular data for a given data set. We will then push this data automatically into a pachyderm cluster.
This will result in a new unique ID for only the molecular data and therefore a new unique ID and version pharmacal set about is that it will not reprocess any of the other pipelines that are required to build a form of asset which comes in handy. As we don't always want to reprocess, Pipelines that have a heavy competition time such as the drug response data. At the end, this pharmacal. Sad version will automatically gets pushed to do, which is an online data sharing
repository, where it is assigned. Toi at the end, the custom key set with Associated DIY chimney, access an orchestra with a custom metadata page, that is sent to the user via email. Now, nor will show you a demo of the orchestra platform. Hi, my name is Miranda Cano and I would like to go over a brief demonstration of the orchestra platform. Welcome to Orchestra a cloud-based platform that provides a transparent, reproducible and flexible computational framework for processing and sharing
pharmacogenomic datasets or peace apps. You can find a list of the latest version of besets that have been created by a lab in the list of canonical. PCS page. You can view detailed information about HP set by simply clicking on the main single piece at page such as this one for ccle 2015, provides metadata, and molecular data sources used to create the speed set. you can also view information about the pipeline used to create the speech at under the pipeline tap
You can download the peace it by simply clicking the download button. You can click on the doi to go to this piece at Sonata page. If you'd like to search existing psat's, or request, a customized piece at go to search and request page. You can search through the existing, psat's, by filtering them with a piece of promoters tunnel to the left. If a piece of it with a desired set of parameters does not exist, you can request Orchestra to generate it by switching the request piece at toggle.
You can submit the request after filling out a additional parameters, pset name, and an email address to receive a notification for the pipeline completion. Your request will be stored in the database and will either be processed immediately if the pocket and they're processing layers online or will be processed when delay it becomes available. Once a pipeline request is complete, you will receive an email which includes a link to the piece at page in Orchestra, a link to the Sonata page and
the link to download your piece set. Orchestra has additional features such as there's that metrics and peace at usage statistics. And API endpoints to access information about existing piece sets. To explore the platform, please. Visit ww.w. Thank you for your time. I would like to thank dr. Benjamin 18th and my lab members for their immense amount of support and guidance. Throughout this project. Understand a great way to start us off. Our next talk is from shotta. Thank you. Everyone
is very exciting to be presenting it to buy a conductor this year. My name is Trump up by and I work and Gary Vader's Lab at the University of Toronto to share with you our software Netflix which helps you build an interpretive outpatient classifier, by integrating multi-omics data by using a patient similarity Network. I'll walk you through these Concepts. The context of this work is that Netflix seems to provide software to help clinical researchers and basic research create a risk models.
Risk model can be used to predict drug response, prognostics and risk of heart disease risk and so forth. And a key part of risk models is the integration of different types of data multi-omics but also other modalities, what clinical data, Imaging data, and so forth. What nutrients does has when provided with set of different essays under different data types. Netflix uses machine learning how to allow you to predict patient outcomes. You can have binary out somewhere hidden. We have some at present we support categorical outcomes. Only use machine learning to classify new patients with
accuracy identify which of the data you provided for the predictor was actually informative and making the prediction and help you generalize to other types of data. How do we achieve? So this is where the patient similarity networks coming to play a patient. Similarity, network is simply a network where the nodes are patients and the edge Wade's. Quantify. How similar the patients are for a given data type. So in this example, we've got three cases and three control clinical genomics, metabolomic data, the patient similarity Network for clinical data shows that the
cases in the control room separately classes. Very well but for metabolomic data only the case is cluster very well. So what a similarity and dissimilarity could be as simple as Pearson correlation of gene expression or it could be other measures of whatever makes sense for your application. So what did she expand does is when provided with an erogenous, patient data and some labels of interest for classification, it takes me 10 erogenous data and converts into a common space of patient similarity networks, it didn't
use it as a step up from machine, learning called feature selection, which essentially puts the score on each of these each of these features to tell you how pretty they are old, is Ben Affleck and then, do you think the top features a spinal integrated navigation? Network is created when you can send has been classified based on which group, they are most similar to, and listen to weighted Network allows you to identify predictive search features on my new patients, the message behind Netflix were published last year about a link here and I will make this top
public after the session So in the message paper, we benchmarked Netflix using the pain cancer survival data, so I hear we've got four different cancer types, kidney, ovarian long. And the task was given five different types and clinical variable predicted. A patient is going to have a good outcome or 4. I was coming from the survival time and each of these plots shows you the performance for Netflix, which is a year as compared to other machine learning message.
And we found that in the Benchmark GTX outperforms, most methods, most of the time, there are some exceptions, like, when the svm message support Vector machine, find the really low non-linear separator for some instances, not surprising, because no machine learning method is going to fit all situations and you can concern at the X1 tool in your toolbox for clinical prediction. Importantly when provided the don't make data, nextx provides need of support to create features at the level of Pathways to. Before I showed you that you
take for example, expression data. Can you make a single PSN Station Network out of it. But what nut takes allows you to do is pull definitions from curated, databases break up the gene expression Matrix do one feature for Pathways. What does does is when you now run feature selection, you are scoring Pathways based on their predictive value for clinicals this improves, interpretability, and is a good starting point for rational drug design. Here's an example. That was part of our message paper, the
task is to classify breast tumors as being of aluminum or a different subside. Aluminum pipe is a type of breast cancer Harbor features. And we run this year, the particular does really well at the end of the day and not almost perfectly separating. Tumor is luminol a or other subtype. But this is what you get at the end of the Predator. It isn't just doing well and you don't know how this visualization is calling in Richmond. Now. It is showing you which Pathways and Pathways memes scored well at predicting at predicting the tumor subtype.
So each of these notes, he's a pathway and it is colored by how well, it didn't the algorithm, please add shared genes have an edge between them. And then you can use clustering, approaches to find seems to do. Well, when we do this with Messy actually fine. The password is predictive of tumors are consistent with the known dysregulated, cellular processes related terms. So you can men take this and drilled out and just Pacific Pathways and asked the question, what is it about this pathway? That's making predictions it generates hypotheses for Downstream expire.
I'm at the ex works. I told you already that it uses machine learning and this is what it does. Give him the user provided data GX brakes, the data into training and test part 2, The Marvelous trained on the first partition. And this is a blind side on which it is evaluate. Like I said before, you start out with the data and it all gets converted into Peyton similarity, networks are we didn't go through a step of feature selection, where networks are scored from 0 out of in right. Whatever your kind of, whatever your hand is for your application, like 10 for each of these
tenets of Nations Netflix takes a different sample of training data makes the PSN. And then run the kind of regular eyes regression which tells you, which of the networks have a nonzero wage in which don't networks that have a nonzero weight. Have their Nets score bumped up by one. You do this again and again for different Sam. Reading data giving you a range for S4 features, you can apply a cutoff for your features and features that scored nine out of ten are my selective teachers. Know, you can take your training in your test set, Justice elected features
and make a single integrated patient similarity Network, then runs legal propagation on this network, which basically Walks from you no known examples of classy down to wall test sample, test patients and known example to class b to all the tests and then you are well, which class is a patient similar to tell classes are a sign and the model is evaluated, this whole box gets repeated. This whole process gets repeated a hundred times or however many times you'd like in order to find which predictors Arkansas, which features are consistently predictive
and how well does in generalizing. Types of data, this is how it works at the x's and bio C has a virgin you know 3.11 I encourage you to try it out and give us your feedback. We got Robert functions to bill predictors the simple functions as a predictor, these many fled here are my features. There are functions for organize features into Pathways of multi-asset experiment object. Type to take the user-provided container for different data, type of software, using the are size three package. This allows you to use Nets,
ex22 allies and communication networks, and the network of top features that I showed you. So there are vignettes showing you how to build predictors with multiple data Stripes Park genetic data and so forth. Going to these applications very briefly review it for predicting antidepressant non-response prognosis in, pediatric traumatic, brain injury, and the future were planning to extend continuous valuetina's fight until the Deep learning variation for increased to nobility. Thank you very much without a like to acknowledge my lab. Gary Vader is lab,
lots of people, commuted. Lots of external collaborators from Munich Verona on Creedmoor that you'll see for a good discussion much wonderful. Thank you for a very interesting talk, just a reminder, for any of the attendees that came in late. Please submit questions using the possible whole feature and make sure to note with speaker, your question in your best, you are next weekend. The session is Gabriel. Odom is also doing a lot. Take it away, ain't you. I'm so happy to see everyone here. I'm going to go ahead and share my screen and
you should be all seems slides at the moment. So thank you for joining us. I want to first, I give Cummins and combinations everyone who for surviving a global pandemic as well as possible. So my thoughts and my my heart go out to all of you. I hope that next year we were at, we're able to do this in person. So what I'd like to talk to you today is using pathway PCA, which is a conductor package to perform a global test for integrative pathway analysis or multi-omics data. I left the link to this to the slides in the chat on possible. I'm part of a
lily Wong and Stephen shins Lab at the University of Miami Sylvester cancer center and I'm also I an assistant professor at Florida International University. So I'm going to go over a problem, current approaches and some of their limitations briefly discussed our method and some short simulation results in a practical example problem as we've seen before these two previous really well done trucks from the curse of dimensionality, and we often want to use Matrix decomposition techniques
to Dimensionality from tens of thousands, hundreds of thousands more manageable, a lot of trouble whenever they used techniques without regard. For existence biological information. Such a very commonly used biological information that would be halfway information. And Analysis is a, a wonderful technique that will allow us to combine biological information with the with the data is one that we can use to bridge the gap between Matrix decomposition, techniques and biological information.
So depressing, PC a package we were able to publish on by conductor last year. Antibiotic 2019. Last year we gave a workshop that shows how to use this this package for your own research. So I encourage you all to check out this form of unsupervised approach. So the idea would be that we have for the data and we have our nasik data that we're trying to relate to a clinical outcome. Let's say for instance overall survival or Microsoft Flight instability, something like that.
And so the one platform supervisor approach would be to for instance used as a PCA or some other pathway analysis techniques to find which pathways are related to the biological response. However, the limitation there is it's not that we can't make systems-level it intuition. We can have systems little intuition about about the survival response. Because we can't incorporate this Prince is proteomics. So it's good for one particular platform, but we can't go across platforms. On the other hand.
If we use our to platform Matrix decomposition, techniques, such as non-negative, Matrix, factorization or Sparks canonical, correlation analysis, then were able to dig into how Pathways work together across platforms, which allows us to look at the system as a whole, but it's difficult a weakness. There is it's difficult to incorporate clinical information to to ascertain or or give value to say. Okay. Well this this cluster of genes and proteins that we found it actually related to the overall survival. So our
checks h e r is given to pot forms of data with matched samples. Let's extract using Pathways and I can Jean and then I can protein from each person and then we're going to fix some form of linear model or Cox personal hazards model, depending on the type of outcome that you have is that allows us to integrate information from, there are in a sick day to. And from the estate of, this is a global test because we're testing the global know that neither of the relationships are are there. So this, so that's a beta-1. And beta-2 were supposed
to do this process as some of the supplemental material. So this is that the prettiest paper that I just mentioned. We it was published a few months ago. Is a simulation study too kind of compare how the password PC a global test is doing against other Matrix, factorization techniques. So the pressure at all paper from last year, Compares, a couple of different multi-omics techniques and we took their two best for farming methods of the Matrix. Factorization and sparse canonical, Corliss analysis, and compared to a power and test size. As we see here, the test sizes were all
recently, well-controlled are less than 0.5. The power. What we did was following the example and we chosen a subset of Pathways to treat and the design points were at to change the proportion of features in the pathway that were treated and also to change the treatment effect. So in this figure, The Columns of these four Columns of figures are the proportion of treated features in the pathway from 10%. Up to 50%. And the road, the feature Rose are the strength of the treatment from two times. The standard
deviation up to six times, the standard deviation, the power is on the vertical axis. The horizontal axis are, are the three methods that we can parrots show in the, on the left side in the salmon, color is the global test which is our method in the middle. And the screen color is the Matrix. Factorization in on the right side. In this blue color is sparse canonical correlation analysis. So as we can see moving from the top left down to the bottom, right? So if we move diagonally across this great of figures, as we increase the proportion of a of a features treated and the strength of
the treatment, all of the Power. For these methods increases, which is great. We also have oversee that across-the-board. The global test method is able to have Firepower outperformed, this classification types of this classification exercise. Then the nun in The Matrix factorization and the foreskin on the clinical outcome. With the two platforms with rnac, can we use the Bavarian eye cancer overall, survival from the TCG a repository we're using the link Tomac hosting sites to get that data.
And we performed independent analysis, using a wicked Pathways at, which it, if you haven't seen, there's a link here. There's a, there's a great are packaged on-site job. And it also plays well with the side escape, which you should really check this out. It's good stuff. The pathways were dysregulated for this for this particular for this particular cancer, because sometimes it's very difficult to have peace out the signal. Once we performed, we use the global test to perform
an integrated pathway analysis. We were able to find a new Pathways and, and confirm that these Pathways made sense biologically, and show that these Pathways were indeed related to overall survival for ovarian cancer. And three of the paper, at least of these pathways are very much worth investigating your time. So we have, we shown that the global test is a technique that we can use to integrate different platforms of data with a clinical response at clinical response, can be
classification, regression or survival. And it will use I in jeans and I can proteins. If we're using protein data through these, I can vectors from the pathways for these two platforms to relate the pathways to the clinical response. Now, for future work, we'd like to reduce this limit this limitation that we have on the match, the match samples. And so we're looking to move to Sri plus platforms and reducing the elimination necessary that that we have these match samples. So
this current functionality and the next functionality into pathway PCA version to if you'd like, to see what we're working on now, and I'll leave you with a couple of closing remarks. First of all, I have a lot of people to think too many people to really sit on the slide, but my co-authors they've been Also also fantastic very helpful. And if you'd like to see the related to the coronavirus South Florida, we've got a we got a website called the Miami covid project.com. So stay safe, wash your hands, wear a mask. You
know, all of that stuff, be healthy and I hope to see you all in person. Next year. Great, thank you for being our final speaker. In the session is Smith video. I need to start. Hello everyone and welcome to my talk. Thank you so much to be recognized as for giving me the opportunity to share some of my watch share. My name is Ruth Smith, and I'm currently at the IRS expense for Instituto Nacional. Condescend. Microbial ecology and my research left a focus on tackling climate, change the issues and agriculture like job I using microbes.
that are available for the are open source, graphing library, that is a very important product by puppy that I'm going to show you today and it was great how to fold that application looks like. So that is best described as a framework for building analytic. What applications? Which means that really anyone who wants to explore and visualize their dates, are in an interactive way, can you stash to open their application in a web browser or hosting the server to share with the community for a
while and Tyson? And since last year's also available and are the great thing about Josh is that its open-source and MIT license, which means it's free. I put together a bunch of resources at the end of my presentation for you to check out if you want to learn how to build a ship vacations yourself. And so the power of - 4 RS, really that, in Combined our statistical models and graphics with interfaces that are highly customizable, I'm going to Illustrated later on, in my demonstration, include
dashboard components that. I'm going to explain in a second and dash HTML components that allow you to give away out to your application. And there's also a few specialized five years, including Dash fire for by informatics and dust stock, which stands for D. Precision and control and includes things like a display for numbers. But I'm showing also in my demonstration demonstrations later on. That's what Ponies are kind of like Waco. Brakes that can be used to build customize apps and to interact with the app and do squares have four sets
of components include things like Ralph, slider about you and the drop-down, or even texting from off of mark down. And so in the next that I would like to walk you through a few typical steps to build a dash application. The first one of course includes loading like the second one is to start your third, one defines the way with some simple HTML in a force that you'd want to Define your component. Said, you want to see in your dash application. So in this case, I need to find a decision. Which
sounds for text with a Bollywood says, hello. And I'm defining a dispute, which is the best to show a bar graph. The interactivity, And so in this call back, I am defining that whenever the value for the title changes, might figure out put changes. The second call back. I'm saying print me whatever I have typed in a text box. And then of course, in order to see the Opera, I have to run the ad on my local server in this case and this is what the application looks like. So this is my textbooks and I'm typing something
into space. Hello by a 2020 Lexus bends, whenever I type, my bar graph is being updated depending on the valley. So I gave before and that's why we can see the print here of what I typed in a textbook, I can do all other things that I can do with a polygraph, including taking a snapshot isolating traces and zooming in, for example. So why does fire we are in the biology conference for developers? You'll know that we're dealing with a huge volume of data and
special formats of data and soul Dash. By really seems to be flexible and efficient. She wanted builder for these kinds of state of it is fully compatible with it and that yard before he attacked another bike and got your package and call my packages that are you are available and can be directly connected with a job application. And are in fact already used by many researchers and buy from additions. Assisted by a gallery and you can head over and check out what kind of tools are already available. And you think with, for example, a
sequence alignment tool and that circular, you know, if you're based on circus and 3D models of molecules example, It's alright, the reason is that put together, feel free to check them out by their share and this wouldn't be possible everything without a bunch of people including Chris and Carson really laid the foundation for the shark. And Ryan were really likes the main people behind of development of our and the whole team of open the open-source development that we can't remember who really
was the main person behind developing - bio and Tyson and let me sum of first lights You're free to reach out to me as well, if you're interested to have more questions. And now it's time for dinner in spaceships. The first application I would like to demonstrate its application of the old for statistical analysis and visualization of my data using 16 on research. And the packages that use for this analysis include packages, like filofax like a bounced back. I draw for a network analysis and visualization and
Counting for Alexander City. And then I spotted the outfit and foremost interactive class. You can explore by hovering over or isolating traces. If you were interested in that particular find them that you want to see how it changes from Front Royal to, and I can have her over the network as well. And knows the pain and treatment and the same for the PS3. If I wanted to interact with this class, I can choose to defend around the ears from the drop-down menus. I can add
another Measure that can breast Simpson and I can change from log scale to win your scale and I can also explore how text, me. Changes over time the same for France with compartment. The second application, I would like to demonstrate a few - bias teachers, and this is an application to build formidable, volatile, organic compounds. That I am starting my research a lot. And so, I took it to work flow. Once it's also a principal analysis that can be visualized
in the sea or you are all very similar West for differentials expression analysis, that shows the magnitude of full change or affect. I can also use a heat map to visualize the values of a compound bow samples. And I can select specific that. I'm interested in going to stop by Mandy's masturbating and then on the day I selected for treatment. And Frappuccino, a few extra settings where I can set my effect size. Excuse me, or I can adjust the threshold and depending on that, this is a very nice. That sounds. So volatile compounds
and biscuits that are significant. I hope I was able to demonstrate a few interesting pictures and show you what as possible, but they are. And thank you very closely. So just remind me to keep the the questions coming and possible. And I want to thank all four of the speakers for four very interesting box. The first question for Anthony is, are there plans to expand its rules of other biological term? A free sample packs to Logic Pro X. so, yes, we have plans to migrate talks illogical data to the platform, in the form of toxico stats, which are beta
objects that are generated from a package that we have developed in our lab called popsicle gx7 expression is for Rada, how is Netflix different from s n f m? And is there a benchmark comparison? Is the phone number for integrated multi-modal data and clustering the samples for classification. So it's a supervised method needs. Labels are related question is it only for cancer? But not necessarily we've done as much control. We did not just in case control with the rare skin disease, Mixel mix.
We have compared nextx is Diablo, which is part of the Mixel mix, sweet and comparison of what. Please come up and performance or what teachers come up and performance for these integrating tools. Performance was gone pro bowl for the Benchmark. We tested for Nancy, XOXO mix with myxoma extends. The score individual features like a Bose micro iron ax is the top that pathway capability is also an unsupervised. Gabriel, how does pathway PCA compared to
Bikes or the PCH Wheels in. Show images of say is this? I wonder if this is a team question, I think so when I last compared pathway, PCA 222 Emoji. I say I saw a really good behavior from it'll. GSA so great work there. I didn't see that. Ours was horribly worse, so I guess that's a good news. To be honest with you, we haven't done a full Suite of comparisons. We were sort of moving into the the unmatched, multi-omics, meta-analysis world. And so a lot of
these methods aren't, they aren't use their this. This result was a couple of years ago. So, when I saw that when I saw that emoji is a talk last year at 2019. I was pretty excited but kind of I thought I was the only person in the space so you should check it out but the other one I heard of that one with the questions deleted after it was the middle of Googling it. I think it's they're not sorry they're not deleted but they it's because there is a ranking because people can vote. So they move up on the list is might be but we have are all questions. Okay, well I
I do not see what was the name of the other, okay. Alright. I have heard of it but I haven't. I haven't checked it out yet so it is now on my list. The next question is for Ruth Given that ass is from flat leaf is that offer different or better plotly integration than Shining. Humane Society and plus when your gas, and what's the difference between a seagrass? So I can only say that it's a very easy integration with any or any Gigi, father can be converted into traffic. Can be directly linked to the callbacks Anthony. Does your tool support beta with
treatments by drug combination, modifying the pharmacal set structure to better integrate drug combination data I don't feature in the network is the feature. So if you have for example, in my first example, you had to you know, clinical gene expression, some other old mutation data, and you make one network, out of these four features, you can take one gene expression Matrix and break it up into pathway level features, you got 2,000 features and then feature selection will score each of them.
So every network is a feature. Can you give a detailed explanation of the simulation data? You mentioned the performance of the global Plastics because it could well capture association between biological and clinical features. But I'm not sure how the simulated data are close to real world. That is a great question. We used to the Future simulation study because it had just been published and at the time it was the the state of the science after working through the data, we saw that it was lacking in
some I guess biological certainty, you know that. So what we did then was the next simulation that were working on right now is a building straight from TCG, TCG, a data. And we're using the word using much more clinically relevant data, to do the simulation bikes at the time, we wanted to stick with what was published to show a reasonable apples-to-apples comparison. It's not It is relatively simple multivariate normal data with the with some other multivariate
distributions. So there's not an extreme level of biological sophistication to that simulated data for this simulation. The next simulation is, I've got one that we we just finished running a couple weeks ago, much more sophistication on. So look forward to that in the next couple of months has made and Son panels be highlighted in others. Walking another. But I asked you about the car on and it all depends on how you just find that in your call back and you can have several input for one called back, so yeah, you could do that.
Anthony. Do you have any measure of data? Quality of the study Levolor cell line within study? How do you deal with data for a sensibly? The same cell line in different studies? Is there any attempt to compare a combined through some sort of consensus? Do not have such measure? However, you can compare unexamined similarities and the differences of the drug sensitivity data for the profiles of the same sound line across a different format. Using the Farm, open
up your package. There's nothing in that. The ex that makes it specific. Patient classification, Sample metadata is an example of classifying cell label. You could try it on. In fact that the X does handle missing data, so you could try with imputation and without imputation of data. So, you know, your your notes could be cell types, or they could be, you know, whatever is relevant for other applications micro classification, for example. Wonderful. So we're at the end of our session, please feel free to follow up with the speakers. If we didn't get to your questions, all the questions
Купить этот доклад
Купить это видео
ConferenceCast.tv — архив видеозаписей докладов и конференций.
С этим сервисом вы можете найти интересные лекции специально для вас!