200: A workshop on discovering biomarkers from high through put response screens
Christopher Eeles (University Health Network)
Petr Smirnov (University of Toronto)
Arvind Mer (Princess Margaret Cancer Centre, Toronto)
4:00 PM - 4:55 PM EDT on Wednesday, 29 July
This workshop will focus on the synergies between analysis results from the PharmacoGx, Xeva and RadioGx packages and their usefulness for discovery of biomarkers of drug and/or radiation sensitivity in cancer cell lines (CCLs) and patent derived xenograft models (PDXs). We will discuss issues with data curation, consistency and reproducibility within the literature as well as illustrate the importance of unified analytical platforms, data and code sharing in bioinformatics and biomedical research. In this lab learners will be led through an analysis for each of the three packages on data provided by the download functions within them. The results of these analyses will then be explored to highlight how drug and radiation dose-response profiles in CCLs and PDXs can be used to discover potential synergistic biomarkers for drug-radiation and drug combination therapies. The resulting biomarkers will be discussed in the context of translational cancer research and clinical applications of genomic data. We will conclude with a discussion of how these biomarkers can be used to inform future in vitro and in vivo treatment screenings and ultimately provide useful insights for clinical trial design.
Moderator: Lauren Hsu
Christopher is working as a software developer in the Benjamin Haibe-Kains lab at the Princess Margaret Cancer Centre where he is developing analytics software, databases and web applications to facilitate the analysis and sharing of pharmacogenomic data for cancer research. Christopher is an honours graduate of the Bioinformatics Graduate Certificate at Seneca College. His undergraduate work at the University of Guelph earned him a B.Sc in Biology with a focus on the biomedical sciences. He is passionate about health care research, with a strong interest in data storage, management and analysisПерейти в профиль
Dr. Mer completed his PhD at the Max Delbrück Center for Molecular Medicine and the Free University in Berlin, Germany in the field of bioinformatics and machine learning. His research focused on developing machine learning algorithms for the prediction of protein subcellular localization. After his PhD, Arvind joined the Karolinska Institute, Stockholm as a postdoctoral fellow. During his time at the Karolinska Institute Arvind worked on breast cancer and acute myeloid leukemia patient genomic and clinical data analysis using machine learning methods. In his research work, he analyzed sample size and sequencing requirements for clinical implementation of sequencing based molecular subtyping of breast cancer patients. Combining his previous experiences in machine learning and cancer genomics, Dr. Mer's current work is focused on the prediction of molecular markers for therapeutic resistance in breast cancer.Перейти в профиль
Okay, hello everyone, Welcome to our workshop on, how to put butt drug response, screens. So to begin, I just wanted to point out a few things. First of all, if you were going to be working off, primarily pre-built package down, been yet that's available through the link here, but being an R markdown. As with all the other workshops, if you go to the link at the bottom of description, you'll be taken to an art studio. Session where you can follow along as well and
run the code as we're going through it and just to find the VIN. Yes, you can. It's in the vignettes folder under Jack's Workshop. Primary reason why we're not working through the our students, because the figures end up quite small and hard to see. And then, as Levi mentioned, there is during the workshop right now. Live help desk. If you have any issues, any are bioconductor issues or you want help running some things. And the link is also in the chat box
on the path of website. So then, I guess we should get started. So I'm going to introduce myself. My name is Peter. Smirnoff. One of the IMO HD candidate at the University of Toronto, and I'll set the Princess Margaret cancer centre in Toronto. And one of the developers of the package that we are going to be discussing today and then I'll let our van and Chris I introduced themselves. Hi everybody. My name is Calvin and I'm a postdoctoral fellow at Princess Margaret cancer centre and I'm one of the dollar part of the package,
do you take the introduced? I am Chris. I'm a software developer in the Benjamin. Hate games Lab at the Princess Margaret Cancer Research Center in Toronto, Ontario. And I am the primary developer of radio Dax. And I also work with Peter on pharmacogenetics and some of the other GX packages. okay, so if you want to get started with the introduction, Yes, okay, so welcome everyone. As you might know, the title of this Workshop is biomarker Discovery. From high to training data set
is mostly focused on how do we find out how it will try to find biomechanics biomarkers for the drug responsibility in by Marcus from the height of its cleaning.. So as you might all know that in the case of Cancer, pass the medicine but is requirement to find the biomarkers and we try to connect with the drug response, response can be from the patient but Finding fish and daughter on a clinical. Trial data is very expensive and having the skin contrast is
mile marker. 100 systems are satellite they have in the workhouse or cancer research. So, basically $2 back and canceled line to take a patient sample. Get you a sample and try to grow it in a dish. Not every sample goes in the dish, but some of them sometimes, some of the cells have properties that allow them to grow in the dish and then you can pass them. If it's a line is really stable, you can submit it to 80cc webcam ready. Can order this Airlines And then you
can grow and expand this Alliance in your lap and to do the Ombre cuz you know, mix your question test different drugs on the sidelines. So basically you just different concentration of drug and you see how many cells are surviving after a certain amount of time. So, this is a normal screening. Alignvest, where do you have drunk concentration on the x-axis and y-axis? You have the Y ability, or how many cents are surviving Pacific properties describe such as I see 50 or UCA under the cover of the
curb and so forth. And also, you can do the genomic profiling on the sidelines of this stuff can be automatic musician, copy numbers and how it is to connect these two pieces to the genomic data. Along with the drug, respond started to find the biomarker, Response from the genomic features so that in future if you have a patient which have just bought the property, you can recommend at least that's the hope of the position medicine, cancel this procedure to streamline, all this procedure to have the gentleman that out very well formatted, as well as the response that are available for me,
that we have a lot of different kind of analysis that we are going to talk about another aspect or another. This is called Regional next ice cream here, you have to consult lines, you Best you expose them to different amount of radiation. Here in the XXX, you have the doors of the radiation. And you look at the, how many, or what percent of cells are surviving October time. And again, I am here is to connect This Time by Marques for the radiation, This was the in vitro, cuz, you know, mix because we are working with the models that are grown in a, in a dish or
the people are trying to more and more juice in Vivo kind of model. So, what model or patient drives? You know, that's what we do is that you take a patient samples, give us a millennium bonded in England. Efficient miles is dis mices in your deficient, at one point, when the tumor is large enough, you remove the tumor removed a tumor and you can plant it into the numerous. This way you have more than one person material material. At one point, when you
want to test your drug, Nigel kiss, let's say you have six p, d x s. You declared three of the control. So they do not get any drug and the rest, you give your choice of drug and invest case and every major the T-Mobile Across the time or place it every week, something like this, when you have a control and the treatment go in the control group that increasing with time while in your treatment group, the volume is decreasing with time if your drug is effective and same as previously, a mention of your name
is too, the genomic data from the patient. All the PDF with the strongest van starter and to find the biomarkers, which contradict the struggle Sports Okay, so after next, I'm going to talk about some of the packages we built for handling, these different types of experimental data. Those packages are pharmaco GX470 Farm, a good cochino radio, GX for in vitro radio, genomic data. And finally zeeba for in Vivo. So we'll talk a little bit about the data structures. They provide
and how to access information within them. So we'll start with pharmaco GX GX provides the farm, a asset class which is a standardized container for drug screening and molecular profile profile data. So if you have a ticket, if you take a look at the class diagram, I'm available in the pharmacology. Exit section, you'll see here each box represents and Within the pharmacologic class. So you'll see Farm Cosette in the Center and surrounding that are each of the sub components that make up the farmer Coast. Within each of these
boxes, the first cell indicates, the name and type of the object. The second cell indicate the object structure on what is composed of and the third sell indicates, the accessor methods, provided by the pharmacogenetics package for getting data from each of these subcomponents. So, as you may have already noticed, there are three major categories of data stored within a pharmaco set object. Firstly, we have bream our metadata in a patient's second-lien blue. We have our molecular data on, in this case
stored as a list of bioconductor summarize experiment objects. And finally in bed we have our treatment response to data from the respective drug screenings. So I encourage everyone to explore this diagram, but in the interest of time, I must move on. So I'm going to go set objects are in general too large to be included in an AR package. And as a result, we provide a set of standardized funk. I'm for accessing information and downloading these data sets online. So the first of these is the available psets
function time and this is called with no arguments and what you get back is a dataframe including the names of all available tea sets as well as metadata such as the associated Publications, type of experiments and the DOI of the farmer Cosette object and this is actually related to both the identity and version of that Franco said object. Once you have selected the pharmacist said that you would like to work with you download peace of function of its first argument, is the piece that name as
an optional second arguments. You have saved directory and this is just where you want that file to be saved, calling this function, Returns the object. And you can see here, we are. Variables corresponding to the names of these respective pieces moving on to radio GX. A radio sets is very similar to a farmer Cosette difference. Being that it's stores radiation response data instead of dose response data from. So examining the class diagram in the section, you'll see that it in yours. The same patterns has Cody X-Men. So I'm going to assume that
from our last section. You be able to read this diagram as well, moving on to accessing our sets objects. We provide and now I just set of functions in the radio Jack's package, namely available are sets which will get you a list of all available are sets. At this time, there's currently only one implemented. From the Cleveland Clinic in order to download the subject of the art that you want to download and an optional directory to save that violin. Moving on to the Ziva. Package
object about treatment, response experiments primarily in patient Drive scenic route models somos, model with cancer and as a result of the difference in experimental design between India and in vitro experiments, the object structure and its successors. I'm also tend to differ from pharmacology ex and radio TX. So, having a look at the object structure. The first slots to within the object is the experiments. Lot, this Lot stores. Need a list of PDX model class objects,
which store has the treatment response, as well as numerous metadata, and annotations related to that specific model. The next major concept is important for understanding as he defect is the idea of a batch. Now, if you look at the figures are available in the Zetas that section, you'll see a diagram outlining that and a batch is essentially a collection of PBX models from a single patient with a single treatment and within Each of these patches. You have two branches, your control branch which receives no treatment.
And you're keeping branch which receives some drug or combination of metadata associated. With the different models. Can be accessed in the model slots lemon. We'll talk about some of the methods we provide for accessing the state at a little bit later on. Finally, analogous to pharmacology XM radio. TX you can access molecular data associated with the azita set in the molecular profile slot. And this is also named by mockler datatype but uses an expression set instead of a Samurai's experiments.
So downloading is a little bit different from the other two packages as well as we have integrated be available and download functions into US, single download music. I'm so when you call this function with no arguments, I'm just like available. If he said or are set in return dataframe with the names of the available set such as well. As many as those in, when you're ready to download one, you'll just passing the name of this. Even said she want to download. So for this
Workshop is actually included all of these dataset. I'm so to load the associated ones for pharmacology X, you're going to run data gdsc and data CPL. Eat both of those are all caps for radius at. You're going to run a data Cleveland that's with a Capital C like the city. And finally, to load the data for Ziva. You'll run data brca underscore. I'll lower case and for the rest of the analysis. We also need to assign that to a variable called brca. So moving
on, I would like to just a question about before you go ahead. So as you talk about this available reset and download the Talking to Jim and the genomic data for The Sims Online, I need to come in and Yes, definitely. So basically what we did was or what we continuously do is we look at the largest Barbacoa genomics in radio genomic studies that do high-throughput screening combined with molecular profiling, such as the CC Lee, the cancer cell lines, liquipedia the singer project genomic subjects of study and cancer together. Now, they
form the depth map project, but also other studies as well. And what we do is we take the data that are released either through their own portals or through Associated files with the article and we process them into this packaged update. A structure, built on top of the bioconductor classes and make them available for download. And we're using. So you can always get apologized to. My screen is a bit narrow because I'm zoomed in but you can get the associated publication when you run the available piece at function or are set and then Where what were using right
now to get. Make sure that each dataset inversion has a deal. Why is where uploading them to a service called to know do which is like fake share or other data repositories? It's meant for ensuring that the object has a deal. Why? That's and you can always download the exact same version of the data. and if you want to know a bit more about how we're doing the pre-processing of all this data, there's actually a short talk tomorrow and at the conference by Another member of our lab. I'll post the link to that Anthony.
In the chat box once the workshops over and also on the slack. I hope that answers the question. Okay, is there is no more questions. I will move on to the next section and mercy will be, if you are following along into our studio, I want to come and that we provided some data stats, for the purpose of illustrating, the functions in the, in the vineyard package. But they're usually slimmed-down version of the data again just because Are package of several gigabytes is not very user-friendly.
Okay, so now we're going to talk about examining and extracting deed of Interest. So we got an idea about the data structures that are provided by these three packages. Now, we're going to look at how do we get to that data in a reproducible way? So to start talkin about pharmacology XM, radio GX GX, a dependency for both of those packages, what does package does is it provides shared infrastructure and allows extracting. A lot of functionality from the individual packages into Corgi X-Men, this includes defining a
class the corsets from which the radio sets and pharmaco set both inherited. So, as a result of this, very similar accessor functions, those two packages, I'm so for accessing data within both kinds of objects. So from here on out I'll refer to them as coarse as that refers to both or a radio set. We have a number of accessor methods available for the different types of data that I discussed in the class diagram. So firstly, we have metadata and Within These objects. There are two common that it
is lots, that are useful directly in analysis. The first of those is the sell slot. What stores sell online imitation for either a radio said or and this can be accessed. Using the cell info method passed a corset object. And what it will return. If a date of frame of a cell cell line annotations with to standardize columns, the first being cell ID on which is a unique MC rated identify of the cell line and tissue IDs, which is a unique and curated identifier of the tissue. In addition to this number, a variable number of additional columns will be returned with correspond. You selling
annotations from the publication that 50 dataset was constructed from. They may contain information in some cases. There's only a few columns are fairies the next. Major meditative slots is a spot containing the treatment response data for the experiments and those differ currently between the two packages and for accessing the pharmaco, a response in a farmer Coast. Access to drugs lot. So analogous to the cell info, you would instead call drug info on a piece set and get a data frame with the annotations associated. With the different drugs, used in
the farm cassette for a radio set you accessing the radiation slot and correspondingly. You call radiation info to get the amount of data has lost most of my data within a radius. So within both of these objects, this is stored in the molecular profile slots does it is a list of summarized experiment objects named by the type of data that contains for example or an acnb mutation excetera? So one of the most important accessor functions to know when we're dealing with molecular data and called MJ two names. And what this function will do
is it will take the your corset and return of character Vector of all available data types. These can then be used with more specific information from each specific. Summarize experiment object contains within that pharmaco set or radio sets. So there are several common accessors that you should know about vs. Spino info. That'll take your object as well as a character actor specifying, the the datatype you'd like to retrieve and this will retrieve phenotypic. Meditator
on the samples in the given summarize experiment object. Taking the same arguments is the feature info function and this returns Need a frame. Containing feature, annotations and metadata from the summarize experiment object. And finally, to access the raw as a measurements, you will call the molecular profiles. Accessor methods in. This will return a feature, buy sample Matrix, where the values are the actual acid measurements for each bullet Define all data type to discuss.
Accessing is the human response data stored in both the sensitivity and preservation. Lots in this case because sensitivity is much more common and I believe is currently only available in radio sets, we're going to use these. As the examples. You can take any of these methods and replace sensitivity with a Shinto associate Access Data from the perturbations lot. So just start with the sensitivity real function and what this does is it going to rain in
three dimensions? The First Dimension represents the cell line drug combination used for that treatment, the 2nd Dimension represents the dose level, given to that drug selling combination and the third dimension in the fur. Send text stores, the drug concentration level, and in the second index store, Steve viability data for each dose level. In addition to the raw, sensitivity data, we provided the sensitivity profiles method which is there to return a set of precomputed, sensitivity measures. So for example, in a farm a asset we have free computer things like ic50,
AC and some other standard sensitivity measure it for you so you don't have to run the functions which can sometimes he computation intensive finally to access metadata about the sensitivity experiments. You call the sensitivity function on, either a piece, a corn are set and this will return dataframe with metadata. Okay, so I guess we're going to talk about substituting pharmaco set object. And these are equivalent for an art set except in all places where we specify drug use switch to radiation and there's two major modalities for such subjects.
The first is using this method and named Arguments for example in Pharmacology X drugs. There's also I believe in argument called cell which will allow you to select cell line so you can see her in our example of a data set and this returns as summarized experiment. Containing only sell lice treated with that truck. the second modality of subsiding in these objects single bracket, Can be thought of as a cheetah mention object, the First Dimension being cell line and the 2nd Dimension being the respective treatment either drugged
or radiation on. So you can see it. Our example here, our subsetting pdsc for the Whitey cell line, treated with paclitaxel and it Returns the associated subset sum ice experiments. There are some exercises that we provided here to work through. We recommend that they are fun and a learning experience, okay? If I could just not depend on cord e x, yes, we'll see what happens in the future. As a result of the accessor methods are quite different. Some of
the key accessor methods for this to get metadata about your different PDX models available within that object, you Call of Duty model info function. On the associate at CVS at this will return dataframe with the model ID, as well as a number of authentication. Problems associated with those such as the patients came from with the tissue type is the drug treatment. If any Next important access, our method is getting experiment and this will return the actual treatment response
data for a given model. So you can see here were calling get experiments on the brca datasets for the model. ID of I won't read it and what we learned from that is a frame with the model ID, the drug name as well as the associated treatment response data, Final important message, message matches. As we mentioned above a batch corresponds to a set of samples from a single patient with a single treatment and it has both a control arm and a treatment harm for whichever drug or drug combination was administered. So
this method actually functions in two ways If you call just Bosch info on a 06 Spectre of available batches, if you know which batch you want to access, you can get additional data bypassing in the batch argument specifying the name of the bachelor want and that will give you back a list, containing the name of the batch, as well, as the name of, what's the control and treatment branches of that batch, which you can then use for finding the different models associated with us.
Okay. Arvind. Is there anything we should address? Right now in the questions. Okay, then I'm going to switch gears a bit. So we talked a lot about the technical sort of, how do you download data, how do you access it to extract it out of our objects? Now, I want to talk about some of the functions that these packages provide to visualize the data in these objects and also to start the model. This sensitivity data and try to derive some summaries, some single basically take a curve and drive a single number that you can then use to categorize the lines, for example of sensitive or
insensitive to treatment So starting with Pharmaca GX with drug treatments and plotting the data. So we'd have been prevented a function, we call a drug dose response, curve trying to be descriptive of what it shows. And what it does is it takes sells soy cell line names and a drug named Karen putting lapatinib on these three cell lines and what I'm going to do. So first of all, these drug names if you want to be sure exactly which name to use for the drug depending on the data
said, if you called the truck, the names are look at the drug info slot. It'll show you what the stranded, ice identifiers are in the data set. Under drug ID and then probably drug dose response curve. I can pass in the name of a cell line here. It says came L2 and lapatinib looking at the response and I can pass in one or more formal cassettes to create a plot which is just do Saint the log scale versus viability and it applied both of these together and also show you the
overlapping concentration range and we have a few examples in this. But yet, I believe I apologize. One of the legend labels isn't plotting, just because we subset of the data and we lost one of the columns for the area above the curve. For the gtac, date of said, why don't you have the full data said? It should pluck the both the labels. And as you can see this, let you look at the actual Rod, screening data and to text things like when two data sets
disagree, hear one of them shows no responsibility. While the other one shows increasing response with increasing dose. This same function can be used to plot your own data here. I just create some very artificial data and acid into drug dose response curve using the concentrations and by Billy's parameter. And this takes a list mainly because again, you can plug multiple curves and it plots it for you. That's how you visualize this dose response data and then One thing that we might want to do is so sometimes these curves are quite
clean. This one is a fairly good curve. These ones have some noise in them, but a lot of the time you want to fit a theoretical model to this and then try to drive parameters, such as the ic50. Or what's the estimated concentration? That's required to reach a 50% probability. And so for this, I'm going to assume in just a tiny bit GX, we implemented a function to fit a, he'll Curve Model to this data. The hell Curve Model basically is a log logistic model. So it's a logistic model. When your concentration is on the bug scale, it takes this form. And
then it has three parameters, one of them is the Infiniti or we call the infinity, it's the maximal inhalation predicted with an infant construction of the drug. There's the ec50 which of the concentration at which you see the inflection point of your loaded and there's a hillslope parameter which controls the slope of this curve and also can be interpreted as looking at whether The Binding of the drug to its Target is cooperative or I guess weather. Inhibited, and whether one
bound drug can inhibit further binding to other targets. And so a, he'll curve usually looks something like this. And hear the ec50 is the inflection point. While the ic50 would be where it crosses in our nature to 50%. We implement the function called languages to progression again, given values, it'll you pass them in and it gives you back the three parameters. And that. But for basically simplifying users analysis for every date of said we also pre
compute, these parameters and store them in the sensitivity profile spot. And then we want to summarize these curves often for analysis into a single number, and we Implement several ways of doing that. One of them is the area above the curve, which ideal see that throughout the vignette is our preferred method. But again, as I was discussing, there's the ic50 which is the crossing point of this 50%. The inflection point could be an interesting primer for you, or the emacs, which is the maximal observed inhibition at your highest concentration tested. And are these.
We also provide functions such as compute AUC or computer ic50 for you to be able to easily compute these values. I'm going to move on to radio checks. So it's a very similar idea. You also have dose verse response but actually the model that you fit to the state is different and that's why it exists in its package. The standard model used to model relation. Response is actually linear quadratic model which is an exponential with linear quadratic equation.
I apologize, this should be a plus there. The way you interpret, the two parameters, it has an alpha and a beta parameter. The beta parameter is a measure of how sensitive whatever if you were profiling was to double hits of radiation. So two events of ionizing radiation, causing damage to the DNA double-stranded break, and the alpha parameter is sensitivity to a single damage event. So yes, in the vignette, we go through. Extracting, the Ralph sensitivity data from our
study and then using that data using the quadratic model function, 252 parameters. And it also gives you back an r squared measure of goodness of fit. Similar to pharmacogenomics you might want to summarize this into a single number and two summary measures that are often. Or we found often used in literature, are the survival fraction at a dose of two gray, which is basically if we go to Sorry, I'm just going the wrong way, at a dose of two gray. It's what fraction of cells survived. Or the
d-10, which is a sometimes extrapolated measure, but at a dose of 10 gray, what is the predicted? What is the dose that is predicted to leave? Only 10% of the sole surviving So it's the you find the 10% Mark and it's the dose that intersects that and as you can see, we also implemented a curve to put this data. We just call it dose response curve and given values in a plot the data and if you use plot typo that'll actually pluck the fitted model as well as the points ending to Legend, you get all
the information about your car. I'm going to move on to ziba arvind. If you want to take over these lies the data into out which device you were selected to look at the individual CDX model. So the first function is blood PDX or here in this function, you provide the object name or your name and the patient ID for which patient, you want to see the model and the name of the drug that you want to see, and it will block the response of this particular drug on the PDX. Go here, when you see the volume across the y axis and the time across
the x-axis, But if you want to also see the control PDX along with this, then you will also specified the name of the control. So he had the control is called Uncle name. Also the controls to hear the red blood, red line to control PDX and the blue line represents the treatment idiots, we have only one for 1 PDX in the control arm, and only one PDF in the treatment. So that's why you are seen. Only one line for treatment and control, but if you have multiple models, you can see individual also.
So, this is this particular object contains replicates is dotted line. Represents One X and the solid lines, represent the air PDX models. It's so this is if you can also raise realize this. I don't suck normalized volume. That is the way he described the initial and /, initially volume of this, just by mentioning volume. Normal. If you specify this option than the blocks building on lies in, considering the initial volume is zero, in this case, if you see the PDX car is going down on the volume is less than zero. It means you're seeing a response. So
you're thinking in the volume Some people like different kind of videos from can provide a lot of options to visualize the PDF. Don't. In terms of quantifying the response as we have discussed earlier, her insolent Behavior, caracci C15, all this message via mattresses for the PDX. Douglas one Constitution also, And many of these mattresses are already be computed with the object as well as we have functions to compute these mattresses, we can compute the function to compute. The response is called simply
response back by the object and the computer and what make a list of mattresses that you can compute. So he'll be do something which is what if I used in the clinic. So this gives you different kind of information about December 6th, including the changing volume response, so forth and running as a PR Isn't it gets a really detailed information about the competition and all the angle between taken control and so forth. Know if you have notification or you are doing large-scale experiments, illusion love is King daughter. Then you can visualize this. I
simply using the flat and dysfunction so fast to extract, all the response of images based on using the Samurais response function. So the summer is response function, the lights you response from different PDX models and good standby patient or bite and buy drug. They get off your. All the experiments on the previous models and then plot and registration Netflix. And you specify hear what is the name of the control that you can put it in the What you see in this heat map is the response is in terms of em resist as we discussed earlier which is divided into four categories
Tiaras for Computers phone, respond, stable disease, progress of DC. So basically blue means you're pregnant. You're not getting here represents PDX s from one particular patient, and the drugs that have been tested on. Those PDX has a stone as of your name and the first row, the presents, the untreated other controlled experiments that you're doing if your TV exist until trial example, this particular patient in the middle, you see them a lot of blue boxes
Genesis of this parking location. Box wine for some of the patients, like they're almost at the end. Most of them are red or yellow. Just a check for questions. Will I scroll? Chris, was there anything? Now there's a question about generalizing our packages to non-cancer uses but I think that's for that we are running close to the end. So I'm going to stay, only a few quick words on the summary functions. Basically, these functions are implemented. When you
have replicated molecular or sensitivity screening there, just a convenience function, to build a matrix. Such a one percenter line. You have one column for drug. You have one row and They do all the averaging, or you can choose various ways to average measurements. And then, they also fill in any missing values in these tables. So, we have an example here of how to look at across all the cell lines and are provided data sets response to different drugs. but
what I want to get to Is the section on drug sensitivity signatures and also leading into biomarker Discovery. So, in our packages in Pharmacology X, we implement the function called drug sensitivity Sig. Which what it does, is it implements a very sort of first-class analysis of the data and it looks at linear associations between each molecular feature that you provide to the function and are examples of yous are in a, but it also use can be with used on buying her eyes, mutation data or
on a copy number variation which in that case we'll look up log are ratios. And here we Show an example for the gtac dataset, and using the area above the curve measure. So this is the area above this drug dose response curve for each experiment. Looking at a few drugs and a small subset of the genes, it'll estimate, it'll give you back a standardized Testament of how strongly these to correlate, and also some statistics, and a P value in an FDR measure.
On how significance this Association is given the data that was profiled in the data set. This is done. Similarly in the radio, gx4 read sensitivity Sig follows a very similar pattern and then we also Have an example here for the using. This doing this, possibly manually using Ziva. But again, Ziva also provides a similar drugs since the basic function. So what I want to move onto is our last section which was showing an example of how you can use these packages together. And in our van yet, we have two examples. This is where we encourage you either to follow along in
the rstudio server. That's provided right now or later. And here first, I'll let Chris go over the section, but we look at comparing sensitivity signature between radiation and Drug response. Using the functions integrated into our development are packaged in both radio Jacks and pharmacogenetics. I'm so as a result of this a natural question is to ask how a signature for wants to gamma radiation will compare to Signature the response to drug using this information. Hopefully, we can generate I bought this for
a combined therapy or to gain insight into the mechanism of action. Snow in our examples here. We just wanted to have a note that due to the file size limits of this computer to bunch of these. And you can see here that we start off by loading, the two packages and downloading the appropriate are sets and peace at objects. In reality for this, you can just load them using the data function. I'm so the first thing we want to do is actually compute our signatures. So you can see here for the first one, we were looking using the feature info
method on the Cleveland. Our sets were RNA and we're extracting genes only protein-coding genes I'm so we do something similar for the TBM scdss and we select a number of drugs and paclitaxel as well as cisplatin and staff Robin. And these were selected because they represent both traditional chemotherapy is, as well as more targeted therapies. So, here we apply the radiation sexual dysfunction is quite slow. You can see you're using 16 threads for computation, and we're passing the necessary information specifically, the features, we want to do the computation for
similar thing. Drugs selected in Pharmacology X as well as the features selected in this case were using a c e in vitro pharmacogenomic models and they can be loaded using data GPS e-cigs. And so, the first thing we intersect objects down the only shared. So now we go directly into comparing the signature which wraps around two methods. There's an interface that expression analysis. If you want, you connect your sensitivity signature to a disease, were also influence. genome-wide weighted
correlation which is where we're going to be looking at in this case, so if you look below here, we are applying the connectivity score function using both our radiation signature and our Pictures from Amarillo TX to correlation and we choose a number of permutations that we want to do this over. Now we get to the will you have is your connectivity score a measure of The correlation between response and the selected features that we subsets on value associated with each one.
In this case, the interesting results is cisplatin are the score is -0.51 and this can be interpreted as showing that these two correlations with response are actually targeting different teams within the biological systems and general negative signatures of correlation will be more interesting. Cisplatin is particularly other studies increase accuracy of radiation treatment into my therapy, so the negative correlation can be naively, interpreted to predict that the radiation.
Just flatten Target difference of populations of cells within a tumor. But pictures probably not the simple. Given the radio census. Sensitizing property is known in Latin by its ability to an adjoining in DNA repair and that's in cells. That lack a pathway already sent to be hyper sensitive to radiation. So, therefore, this is likely a synergistic mechanism and not just results of interaction or heterogeneous shows how you might identify potential synergistic therapies.
I think we have 2 minutes left. We're actually overtime strictly speaking, but I wanted to ask if there was a commonly afforded question. Okay, there was a question about Wi-Fi. Michael Jackson used to be cancerous Pacific. So, yes, there definitely isn't necessarily anything in our package, that has to be cancer-specific. We work for a Cancer Center, so that was what motivated us to build the package and we just find that in terms of The diversity of models available, especially with different genetic background.
Cancer cell lines are one of the most diverse options for that to do this. So the research, but definitely, it is possible to generalize this to other disease types, that may have similar experimental designs. And I do want to point out, there's a second case, study in the vignette that, unfortunately, we don't have time to get to, is it easy to create your own Ziva, object or event? I think you can comment on that, but the answer is yes, the short answer to create as
well as the reserve in the package which lets you know how to create it. It's pretty easy to do. You just have to clear the data frame, which provides turn media, volume and time. Columns and it creates the USA. So it's possible, you can always contact us if you face any problem in this matter, if there's any other questions, we will be on the slack and able to answer them. Thank you, everyone for join.
Купить этот доклад
Купить это видео
ConferenceCast.tv — архив видеозаписей докладов и конференций.
С этим сервисом вы можете найти интересные лекции специально для вас!