 That's cool, too. Our next speaker is Louise Hukimayers, also from the Lever Institute for Brain Development, and she'll be presenting on the identification of total RNA expression genes. So once again, the importance of controls and standardization in the estimation of RNA abundance and heterogeneous cell types. Welcome. Thanks. Hi, I'm Louise. I'm a staff scientist at the Lever Institute for Brain Development. And today I'm going to be presenting our group's work on our data-driven identification of total RNA expression genes for estimation of RNA abundance and heterogeneous cell types. So the motivation for this work is that our group is working to study different cell types in the human brain using SMFish with RNA scope technology. So we're interested in the cell size and spatial organization of different cells in the tip, across the brain tissue, as well as the total RNA expression of different cell sizes. So RNA scope is great to measure cell size and spatial organization. However, we're limited to measure total RNA expression because we can only measure four genes at a time with this method. So how do we overcome that? How do we measure total RNA content of a cell if we can only observe a few genes at a time? Our solution to this problem was to develop a TREG or a total RNA expression gene. So this is a gene that we're hoping to find whose expression is proportional to the overall RNA expression in a cell across different cell types. So in SMFish, we're able to quantify the expression of one gene by counting up the puncta or points of light that we can observe in the image. And then we can use that to estimate the expression of that gene and then moving on to the RNA content of the cell. So that's kind of the framework of the project. So we developed a process for completing data-driven TREG discovery in a single nucleus or single cell RNA-seq data set that corresponds to our experiment that we want to design for RNA scope. So the first step being that we wanna filter out the top 50%, we wanna filter to the top 50% of expressed genes. So this helps us find genes that are going to be easily observable in our RNA scope experiment. The next step is further filtering. So we wanna filter out genes that have high proportions year expression. So this helps us reduce some of the sparsity that is common to a single nucleus and single cell data sets. And as well as helps us reduce ties in some of the noise in the math in our next step of the process, which is selecting genes with high rank invariance as our candidate TREGs. So what do we mean by high rank invariance? So the first idea here is that we want to examine different genes' expression ranks. So for each cell, we rank the expression of the genes across the count matrix. And then after that, we then observe the stability of the rank across all of the cells for a cell type. So this helps us find genes with high rank invariance or stable expression ranks. So we're looking for genes that have, wait, this is a point. We're looking for genes that have like a tight range of expression ranks versus a low invariance genes that has a wide expression ranks. So we believe that that high rank invariance is going to help us find genes with that total RNA expression quality. So our process we developed for this is again, we first filter out that count matrix to a top set of genes. And then we can find the difference between the expression rank of each cell versus the expression, the mean expression rank per gene. We then find the mean of the absolute differences for those rank those. And then rinse repeat for each of our different cell types in our single nucleus or single cell data set. We then sum and reverse rank those to come up with a final rank invariance metric that we can use to then choose a few top candidates that we can experimentally evaluate later. So to compute those, to kind of like help find those in our single nucleus data, we developed our TREG R package, which is available in Bioconductor. So it has tools that both facilitate the proportion zero gene filtering, as well as calculate that rank invariance, both in a pairwise manner, where we can find the expression rank of the cells, groups, and then finally calculate invariance if you want to explore the qualities along the way or just an express rank invariance express, which computes that full rank invariance calculation in a more efficient way. So to validate this process, we performed an experiment. So our first step was to find candidate TREGs in a single nucleus data set that we have at the Libra Institute. So this is human brain, postmortem, 10x, single nucleus RNA-seq data. We use single nucleus as opposed to single cell as we are working with frozen brain tissue. So this is the trend manor data set and it contains 70,000 nucleus across five brain regions and nine cell types. So we were able to determine a top handful of genes. And then we evaluated those in an SNFISH with RNA scope experiment in DLPFC tissue. Yes, we had nine tissue sections from one donor. So we were able to evaluate three candidate TREGs and a housekeeper gene to kind of compare the different qualities of these. So selecting our candidate genes, we ran our rank invariance process on that data set and we picked AKT-3, arid-1B, and malat-1 is some top contenders. And we were able to observe that they have that narrow expression rank range. And this contrasts to our polar 2A, which is our classic housekeeping gene where we see that more wide expression rank range. So, yeah, we found that those TREGs have high expression rank. And we also found, we also showed that they had that relationship we were looking for where we had a really nice correlation between total RNA per nucleus in the single nucleus RNA-seq data versus the expression of that single TREG in that data. So that was all. And then we moved on to our validation experiment. So with our SNFISH and RNA scope, we got some really nice images of some DLPFC sections and we're able to segment out Nuclei and quantified the TREG puncta for AKT-3, arid-1B, and polar 2A with the halo software. However, malat-1, segmentation for malat-1 did fail as the expression was so high that it just turned into a blob of white light and we were not able to quantify individual puncta. So malat-1, oh, it's tricky self. So then we moved on, yeah, and then further, we were able to quantify these TREGs spatially across the DLPFC tissue. So a quick crash course in brain, or in DLPFC anatomy. So there's like this wave of gray matter here that is more transcriptionally active, it contains our neurons. And then we have these pockets of white matter that's less transcriptionally active and contains our glial cells. So what we are able to observe, both in the actual image, so the green is our AKT-3 puncta over these blue DAPI signals, is that like we saw higher AKT-3 expression in our gray matter. And then this is quantification, this is the results from their quantification. So you can see that it aligns both with the gray matter and where we know that the neurons are located in this tissue. Yeah, so overall, we found that these follow the expected pattern of RNA expression like through the anatomy of the tissue and we had a lot of fun exploring our images and making these plots for this. Yeah, so then again backing up and looking over the different cell types we were able to quantify in our experiment, we found that trikes were expressed in almost every cell. So we're in the 80s, 90s, 90% for finding this, that's a plus. We also checked the patterns of expression across the three cell types that we were able to determine in this experiment. So we had, so following the expression of the total RNA expression pattern from their single nucleus RNA-seq data, we know that excitatory neurons have the highest expression followed by inhibitories and oligodendrocytes. And in our RNA-scope expression, looking at just the number of puncta we could count up in these cell types, we found that AKT-3 follows this pattern of expression really well and had the closest standardized beta across this slope to our single nucleus data. So negative 0.017 versus negative 0.33. So we were pretty pleased with this result as well. So AKT-3 was our top contender and arid 1B did okay. And it would be, we weren't able to quantify polar 2A overall three cell types. Yeah, so in conclusion, we proposed that TREGs allow the observation of total RNA expression in SM fish with RNA-scope technology. We think that the proportion zero filtering in right invariance process is an effective process to select candidate TREGs in single nucleus or single cell RNA-seq that corresponds with the experimental design of the RNA-scope experiment you'd like to do. We also found that AKT-3 appears to be a TREG that's compatible with RNA-skewp in the human brain and we've got future plans to investigate with that gene. So this work is out in pre-print if you're interested and we also have that TREG package available on Bioconductor. And that allows you to apply this TREG methodology to find TREGs in other tissues or experimental settings of interest. Yeah, so with that, I'd like to acknowledge the team at Lieber. So Leonarda Clouder-Torrez, Hirston Maynard, Stephanie Page, Kelsey Mnugomri and Sengha Kwan and as well as Stephanie C. Hicks from John Hopkins who collaborated closely with me on this and our co-authors on our pre-print. So check that out and then feel free to say hi on Twitter. So, thanks. Thank you for a wonderful talk. If we have questions in person, please ask for a microphone. And I don't yet see any questions referred online but they'll probably start trickling in. So I guess be on the lookout. Dr. Wright. I thought this was really neat. I'm totally new to this technique. So could you tell me are some RNAs localized to the point where the punta are indistinguishable from each other? So it's not just that they overwhelm the signal but they actually all locate to the same location and therefore look like one even though there's many. Yeah, I mean potentially that issue might happen. We did do a lot of quality control where we zoomed in visually on these images and like made sure that the results we were getting from Halo made sense. So we were pretty pleased overall with these, you know, AKT-3 aired 1B and the Polar 2A segmentation from a QC standpoint. But yeah, that might happen to a degree where there is some probably some margin of error there. I'm in the control panel bro in that room. If you go. Good to know, I suppose. So I'm currently proposing to swap the.