 Okay, so we have time for some questions. If anyone in the room wants to, we've got microphone there, there's a microphone up here. I've got one online and this one is for Alan from Divica Argoal. Can you explain the difference between spread and full change and what is the balance between the two that we should ideally see and want in single cell data? Oops. Hi, Divica, thanks for the question. So we, the way we measure full change is to actually compute the change in the expression and spread change here would just be the change in the variance of the counts data. Right, so your second question, which is what is the correct balance between the two we should ideally see? So I think I don't know the answer to that question. I think it really depends on the conditions that we are looking at. And I think that's actually an open question. I don't think maybe someone else in the room might know an answer to this. Okay. Questions from the audience here? Okay, we've got one. Thanks. Yes, I also have a question for Alan, very similar to the one before. So I saw that, well, one question is if you take advantage of the association between abundance and spread, like other methods do. And well, a second question is I was curious that for some of your hypothesis testing methods, those two variables were associated and for others, they did not. So I was curious, you know, if you had consideration about that. Right, so, sorry, I don't think I caught your, I got your first question. Could you repeat the first question? Yeah, so it's known that abundance and variants are associated for RNA sequencing abundance and some methods use this to regularize or shrink the estimates. So I was curious if you had plans or if your method accounts for that. Yeah, so I think that's one of the reasons we explored different normalizations was because I think the normalizations in some ways can account for some of these, I guess, you know, like you can call them confounders, right? Yeah, and I'm not so sure whether, again, I'm not the best person to talk about these, you know, like normalization details, but I'm not so sure whether any of these normalizations that are considered in this application do account for the abundancy differences that you mentioned that are known to be associated with variances. Yeah, but I do know, like for example, there are so many other confounders, like, you know, the library sizes. And so I think that was one of the concerns that we ran into like earlier on, you know, we weren't previously accounting for these differences when running the test. So we thought why don't we just try it after we do some pre-processing that accounts for like, you know, full accounts per cell, right? And then, you know, I think actually I went on, I went on the bioconductor Slack channel and asked people what they thought about, like, you know, ways to normalize before running a D test. And I think there was no consensus based on like the people who replied. I have a question for Avi. So in any time if you're going to be, integrating across multiple data sets, how much does it matter that they were done on the same tissue type or the same species? Can you put together human and mouse? Can you take all the stuff that human and mouse and then put it on to dog or to cat? Just wondering about that. Come up here. Yeah, so it's an interesting question. So here it's a little bit different. So we are working on PBMC. These are all blood cells. Here it was easier because they come in from the same tissue. But in general, integrating the data set for functional analysis like these, it's gonna be tricky because even at single cell RNA-seq, we are having difficulties integrating, let's say, mouse or some other, let's say, model of organisms, right? So right now, even with the known technology is difficult. So right, cut and tag is even newer. So we can look into it, but right now I haven't really explored that much. One more question for you. So I understand it's difficult from different organism and different tissue types, but the PBMC for normal donor and the supposed cancer patient is that okay to run in a same experiment, like to see PBMC or how the CD8s are expressing in normal tissue and how the CD8s are expressing in cancer tissues? Yeah, that's a very interesting question. And it's an important one because if you want to study, let's say, newer cell type annotation because of, let's say, disease, the linear just got changed into another kind of cell types, right? So right now we are doing supervised analysis. So I was talking about the reference annotation, right? Where we are integrating everything into one framework. So it depends on what you have in the reference. The ultimate resolution is gonna depend what you have in the reference to start with, right? You can imagine doing unsupervised analysis, right? Where you have all the data set from, let's say, a disease state and then you integrate everything together, right? And then it depends on what kind of resolution you have even in the, let's say, disease condition. So you can do supervised, unsupervised, but right now it's in the proof of concept stage. It's done supervised, but it can be extended into unsupervised as well. And specifically protein, protein information is very useful, which can be used to split apart the cell type annotation. One more question. I will do one more question and then we will take a break. Hi, I have a question for Alan. You mentioned for your methods that depending on which sort of statistic or whether it's a shift or scale statistic you want to test, you have to set the weights and the exponent differently. Is that something that you analytically have to derive or is there some method to guide for those settings of hyperparameters? Right, so thanks for the question. So the short answer is that there are certain special cases where we can analytically derive it. So that's actually the case where the exponent and the test statistic is equal to one. Otherwise, you would have to basically do like a forward simulation over a grid to find the similar choices of parameters over that grid. Okay, well let's thank all of our speakers once again and we will take a 10 minute break and we'll be back for the next session. All right, thank you.