 So we have a couple of minutes for questions for really the three outstanding speakers from this morning. So if there are questions from the audience, please come forward to the microphones. For our speaker, and related to the TCGA efforts, how our speaker comments on the like for data, for sample preparation and data preparation, and how mostly this is like a web lab researchers involved in this. And I'd like to hear the speaker's comments on how computational side people are involved in the sample preparation and data preparation. And also on the other hand, is for when we design new computational software and tools like computational people, and we work on this, like how the web lab researchers are involved in this. The second question is specifically for our third speaker and related to the subgroup as a molecular subgroup of the cancer. And I'd like to know specifically what kind of computational software you use for that. Thank you. So I'll go ahead and take the first question, which was I think how basically the clinical side works with the computational side in this project. So TCGA, the way we run the pipeline is the front end of the pipeline with the tissue source sites. The tissue source sites are first involved very early on when we established a disease working group. So when TCGA determines that we're interested in providing this resource on a given tumor type, we bring together experts that include pathologists, oncologists, biologists that study specific organ sites. You know, if it's thyroid, we bring an endocrinologist, et cetera. So the clinicians come in from the very first step. They provide us all the guidance we need in terms of determining what our inclusion and exclusion criteria are, et cetera. And then that group works with us to acquire the samples. That same group is then brought back at the end when we start doing the analysis. So the TCGA funded investigators in terms of our data generating centers and our primary data analysis centers are not necessarily experts in every single tumor type that we study. So we bring in those external experts back into the loop when we start doing the analytical phase. Now, we also have every center does their own analysis on their own data set also. Plus, we have these seven genome data analysis centers. And each one of those groups actually works with clinicians generally on site at their own institutions to make the tools more clinically relevant for the disease-specific parts of the data that they put on their own specific portal. So, you know, MSKCCCBio, which is also, I believe, having a workshop this afternoon, they have their own computational suite, essentially fire hose at the brode also does the same. And so a lot of these groups also have their own different interpretations and their own different algorithms that they're using. And depending on what the right tool is for the question at hand is when we bring those specific analysts in with the specific disease experts. And again, right now it's still one at a time. For the pan-cancer group, it's a bit more of a challenge because we're going across many, many different tumor types and the clinical questions are obviously much more difficult to generalize. And I'm going to turn it over to Lee for the second question. Just real quick, there's many, many ways to use profiling and make cancer subgroups. So we happen to use a statistical model based on Bayes' theorem where we estimate based on a signature of roughly 100 genes, the probability of being one or the other type. So it's standard stuff. I have a follow-up question for Dr. Stout, too. I'm not sure how plugged into the current regulatory environment you are with diagnostic testing and thinking about clinical trials and stratifying based on genotype or gene signature and how that interacts with new and proposed rules that would need to be in for testing where testing would need to be validated before the trial. And so we're talking about, I think, a real challenge going forward. That's thanks. I'm right in the midst of that, unfortunately. It's a difficult landscape developing the diagnostics and genomics. We're trying to do it for the distinction between ABC and GCB. You could just as well try it for the various mutations. I think the paradigm is that you, for these targeted agents, you need a co-diagnostic that's developed along with the drug, and they then are approved simultaneously by the FDA. To do that, because by definition you will be changing clinical care, the bar for that quality of that diagnostic is extremely high. So it's going to challenge us to go beyond what we consider pretty good in the lab to something that is really going to change whether someone gets treatment A or treatment B. So thanks for bringing that up. And it's not going to be easy, but an important struggle. I think we're going to need to take the next questions offline because we will need to move forward to our next session.