 What I'm just going to do is talk about some of our work and in conjunction in how it relates to some of these ideas. I don't have a specific overall theme to this, except that to attempt to provide comment and commentary with regard to some issues and principles that have arisen in the creation of some of these resources from the Allen Institute. So the first thing that I guess that we did in context with the original Allen brain atlas was that, of course, you needed some landmark, some distinguishing element for which to compare gene expression, some framework. And this was the so-called Allen reference atlas, which was drawn by Hongwei Dong, now at USC, formerly of UCLA with Arthitogus lab, but he was with us for a few years. And with a small team in a very super human effort, he drew this at complete atlas. The notion was, so I'm just going to just tell you commentary, I guess, in the sense that I believe these to be relevant to the principles that Marianne was kind of requesting. They come back to the relationship of neuroanatomists and informaticists. So the first thing was this thing, why a new reference atlas? Why didn't we just use, say, Paxenos, the standards of Paxenos? Everybody would have had Paxenos on their laboratory desktop. And the reasons that we came about that were not because of a disagreement with Paxenos, the atlas, or the ontology, or his interpretation, for which I think that anyone who's been involved with neuroanatomists and ontologists, one of the first things you realize is that it's very difficult to find two of them who agree on the same thing. Further experience with this over the years had led me to believe that it's difficult to find two of them who like each other. But that's actually a different story. We won't go into that for the time being. I mean, I've really heard comments that you wouldn't really suspect in public meetings between leading figures in the field. So, but anyway, so Hongwei Dong, he drew this atlas. It was, of course, it was his interpretation subject to his influence through Larry Swanson and Larry Swanson's understanding of the rat because that was Larry Swanson's primary source of sort of deep neuroanatomic information. And he drew this by looking at, in the classical way, that for many, many years now that neuroinformaticists have drawn atlases. And that's, again, unfortunately, not based on Trigmas comment in the first talk that there would be a 3D system for, in fact, visualizing drawing deep structures, taking full advantage of the 3D structure. But rather, and why is the question, well, the primary reasons for that is the unavailability really of software to handle really large-scale imagery and to navigate that imagery and annotate in 3D. Some of that has changed. Through the efforts of several people, we may even hear something about that through from Yakapo, possibly. I think Yakapo knows about this area. But in any event, the way that they drew, of course, is they typically print large printouts of these sections that are taken. In this case, there were 50 micron sections. And then they pair and stake and they draw things based on, in fact, a lot of additional information. It's not simply, you know, they see boundaries. For example, I can remember when we first did this many years ago in 2003, 2004, I remember having these plates and saying, well, we're trying to redraw a reference atlas. And so I had my friends over at Microsoft Research and Machine Learning Group, and they said, you know, we'll just send us the data and we'll learn the regions. And I said, well, you know, I don't think you can do that. I look at this and it looks really, really challenging. And they sort of, you know, without directly saying, so, you know, that, well, we're smarter than you are and we have substantial algorithms that can do this. So as often happens in academic collaborations, when things don't work out, you just don't hear anything. People don't write you back and say, well, we're having trouble. You know, it never happens, you know. So anyway, but so nothing happened. And, but anyway, Hongwei, he did this. And so what information does he use to draw this? Well, he's using a psychological information, deep understanding of the connections of the anatomy of the pathways that they have seen in imagery. And the challenges that that presents for capturing this in a system which is navigable from the informatics part, of course, are substantial. Well, you need an ontology. And Marianne and others have been really leading the charge with respect to standardization and ontology in neuroinformatics. But it's a problem because it's a type of field where there's so many different perspectives based on different kinds of information and integrating that. So, you know, these kinds of tools that emerged from our choice of ontology, we had a choice of ontology. They were based on Hongwei Dong's choice. It tended to have a little bit more detail in the hypothalamus than it did in some areas because Hongwei Dong was interested in the hypothalamus and that came from influence from Larry Swanson's work. So there's another risk there. There's a risk of the biases of the individuals or preferences in their research. So these provide, we built this system. It was integrated with the Allen reference atlas. And to do so, we wanted to be able to make a system that you could actually take gene expression, integrate it with this reference atlas, read off information. And because one of the first challenges that was presented was as well, we wanna be able to, I can remember Science Advisory Board meeting, seeing that we need to be able to say what genes are expressed in the thalamus, you know, say, or in the reticular nucleus of the thalamus. And it's not a good question, in fact. It's really not a good question because the problem is a problem of, as we know, a problem of specificity. There are too many genes expressed in too many places. It's a combinatorial nature of it that matters. And so you need ways and tools that enable you to make cross-structural kinds of deductions. You need to be able to use correlative search and pattern-based search. These are the ways the things that shed really light on the mechanism. So in fact, what we did was basically build these grid-based methods which were anatomy-independent. And I feel that that's probably a very important thing for the informatics side of atlas building, is to build in some sense, you want to think of the ontology as a type of skin in the computer kind of programming world. Something that, because it will change and people's interpretations and understanding and refinements will change, we need to do methods that enable the superimposition and factoring on of different kinds of ontologies. And so the idea that we hear in the case of the adult atlas, we built the sort of, there were 54,000 grid cells that you could access the information in and you could get our version of the label of what was happening at that vertex or perhaps you could use the waxholm mouse atlas to look at it. And so these methods have been extended and we did different things like that. As we took a look at the developing mouse and in the developing mouse, we partnered with a fellow named Luis Puele, from the University of Mercy in Spain is a well-known developmental neuroanatomist. And the thought was that what we wanted these reference atlases for each stage of the thing. Now he comes from a different school, so he has a different skin, if you will, on neuroanatomy. And it turns out that his approach is what they call the sort of ontogenetic approach basing that the gene expression and the evolution of gene expression should dictate how neuroanatomic borders evolve and how they should be traced. Well, that's all great, but it turns out that the convergent patterns of that didn't completely agree at P56 with Hongwei Dong's drawing of the young adult atlas. And so concessions had to be made for that. As a neuroinformaticist, one had to provide, like Doug is talking about, translation mechanisms to map between different schemas. And so it's not quite sufficient to have just this notion that even though these are consistent, they may not be consistent with another final kind of realization of it. So there are big pipelines, of course. People know that we do these things. You cut sections, you map, you study them in 3D, you attempt to adjust boundaries, et cetera. And we and others did this sort of thing. One of the advantages of the system that I've always found, an application of this I've always found interesting was the use of gene expression data in helping to define or refine anatomic boundaries. And contrary to, that was one of the big things, I think, maybe several years ago, up to 10 years ago, people thought that you'll get gene expression data and that will, in fact, it'll be a brave new world for neuroanatomy. It'll tell you all these new things. The hippocampus won't be a hippocampus anymore, it'll be something, you know. And people quickly learned that, in fact, that that wasn't really true at all, that basically the neuroanatomist, just by essentially looking at histological sections and studying things and making deductions, had everything pretty much right. I think there are small refinements and boundaries that connectivity data makes, that gene expression data makes. There are some new nuclei occasionally that appear. But for the most part, things don't move around in any great way. This is a, one tool that we did to study this was that it was a consequence of having this data mapped into a common coordinate system was that you could now study either correlations of gene in space or, from a spatial point of view, at a given voxel, you could ask, if I look at a class of genes expression at one place in space, how do things vary for nearby neighbors? And this facility, we call this anatomic gene expression atlas, and other people have looked at this. Hongwei Dong, in particular, used this to find some evidence for certain subdomains in C8-1 in the hippocampus. So it was possible to sort of change the ontology from this kind of approach. So a word or so about connectivity, one thing that we just released for Elite recently was this sort of mouse adult connectivity atlas. And this is based on a combination of sort of viral tracing methods and sort of genetically modified, pre-dependent kind of characterized mouse lines. And the idea is that injections are made in certain areas of the brain and a sort of a tracer kind of traces and follows, of course, the external paths. And from there, you can provide informatic methods to segment this, to delineate signal, and to build a, as Parthymetra and other, and people have commented, and that Jay Boland, who's here, I think, in a nice position piece, a sort of mesoscale kind of atlas. And mesoscale has kind of been the focus of area, particularly in the rodent, because it's tractable. It's not really the detailed microstructure for which we know it's so difficult to obtain full connectivity for, and it's much finer detail than imaging, classical imaging-based methods. So again, you need pipelines to do this en masse, to do this in kind of a large-scale structural way and to provide sort of QC. And by doing this, you can get a sense of the connectivity of what's projected to what. And so we produced about 1,300 data sets, which do this, using these projections. And there are challenges with that, too, that are sort of informatically interesting. One is that here is kind of what you'd call almost sort of the naive connectivity matrix. Here, what I'm just looking showing is that I have two halves of the brain, kind of the ipsilateral and contralateral side. And this just shows an injection into some anatomic area and it's targeted kind of projection here. Now, how is that determined? Well, it's determined by taking the data, mapping it into this coordinate framework, segmenting it, and attempting to determine basically which are not so much termination points, but because termination is very difficult because one doesn't have a synaptic stain, but determine that a highway in fact went through this voxel. And if a highway went through, well, then you must have been connected there. So that's kind of the principle here. And the symmetry of this, from ipsilateral to contralateral, is attractive and it does validate that, in fact, the projections are kind of working out right. But the challenge there is that it's not necessarily the correct connectivity because it's based on injections and the injections take up, the virus takes up a bolus of an area and that area may bleed over into several of whoever defined the anatomy for that region. And so the question is, what skin was used for that anatomy? And so subject to, say, the Allen reference atlas, what we were able to do is build a kind of model which would basically take a structure and assign in a form of regression sense what part of the signal came from which piece. So you could do this at the projection, you could do this at the terminus, and you can come up with a model of connectivity which is based on getting the weights kind of correct. And that enables you to do tractography virtually and other kinds of things. So I just wanna spend some concluding moments here talking about sort of common coordinate standards and stuff which I know are really important to some people. They're important to me, important to people that in the INCF digital atlasing group for which there's been a long-term involvement of Jill Baleen and Richard Balduk and the other thought leaders in this area. And I think that one of the things that, and Doug mentioned that, that one of the things that's really important is to what extent can we use these resources and integrate them and make them standards. Standards, to my mind, evolve by, they evolve from usage, right? From community usage, not from mandate. No one mandates, things don't become popular because someone said you must use this, because they're tenable and tractable. So in an effort to expand, in addition to the Waxholm thing, I wanna try to talk about another effort that we've been involved in, and that's this idea of kind of getting a better kind of refinement of the C57 Black. So as part of the connectivity atlas, one thing that we did was we basically took, we used this sort of tissue site system, which was a two-photon sort of slice and dice system, where you basically, it takes a block-face image each section before it cuts. And so these are really the best systems that we have for translating 2D into 3D because you have the basic, you have your topology kind of in the right space. And so what you're looking at here is just the averaging over systematically large datasets from 41 brains up to now over 1200 brains of average using this system. And doing this one actually can see things that are not quite visible here, but upon magnification you can see the barrels and the barillettes in the cortex. You can see details of the striatum and the hippocampus, which in datasets of this nature were here to fore unseen. And so this is, it's published, it's announced in the context of the mouse connectivity paper, which appeared. But one of the things I think is to get this into the wax home standard and to get this related to that, which we know we have the Alan Brain Atlas map, but we're in the process of kind of refining this in a way that I think that would be a good thing. So let me just conclude there. I think that, I mean, we, my comments are meant to sort of just highlight some of the challenges connecting informatics and neuroanatomists and the relationship that the two have to each other. So thank you for your attention. Oh, I did it. Okay, microphone over here. Is this the microphone? Hey, first of all, thank you very much for this very impressive data set, especially the tissue vision one that was released kind of in connection with the last paper. Now, my main question here would be, will we at some point have like a consistent 3D parcelation of this particular data set for as it is with the Alan, with the regular Alan X Atlas at the 25 micron resolution? Because right now, if you overlay these data sets, it's still very much apparent that the parcelation comes from a, from coronal 2D section. So as you scroll along in Z, there's a bit of quite a bit of jittery. Is that something that you're addressing? In the connectivity Atlas. Yes. Yeah, yeah, we're trying to actually produce that and produce a comparable API that would enable people to grasp the data in the same way as the adult. It's a work in progress. It's not completely done yet. Are there any timeline on this? Sorry. Less than six months, I think. So I go, since I've got the microphone. That was great. I loved all those brains that you've averaged together. So two questions. First is what is the actual resolution of those? And then the second is, have you looked at the variation from the brains that came in just to see how much variability there is in the black? Yeah, some of the variation. The resolution is, I can't, I have to recall that. I can't answer your question off the top of my head to give you a precise answer that, and the variation we haven't studied that, although there's been people actually who have, I received an interesting email from people who have kind of liked to pursue that. I think it's definitely worth studying. I think I like about that schema is just that, see, in the original Atlas, it was all about cutting these sections and then reassembling them first prior to doing this deformation step into the Atlas, and this takes away that. If we had had that from the Alameda, brain Atlas, things would have been even better, I think. So I really liked your talk. I like the fact that they're doing different developmental ages. You keep saying the mouse brain. So can I assume that all this data was generated from both males and female animals? And if not, how come you're calling it the mouse brain? And not the male mouse brain, and we don't really know what the female mouse brain does. With all due apologies, I'll avert that question by saying it was only due to the shortness of the program that prevented me from being more specific and accurate. It is only the male, seven-week-old male adult mouse brain, and things might be different looking at the female mouse brain. We will not. We have looked at females in the adult human, but in the mouse, we probably won't go back and do that, and it's not out of lack of belief in its importance. It is more driven by that I don't control the financial decisions of the Institute, as it turns out. And this is actually a question for everyone in this session. So if I go and do some studies in female mice, does that mean there's no point in me even trying to provide my data because there's no mouse brain I can map it to? Because I certainly can't assume my female is the same as my males' institutions look. I highly doubt the answer to that is really discouraging because of the profound sort of homogeneity of these animals and their genetic similarities. I would be very surprised, but I mean clearly one should do those experiments and one should provide that data, but I guess I would be surprised that it all was lost. Yes, it should be possible if you map your data from females to the male atlas. And your system, the warping system, keeps track of the warp parameters. Then you can tell from the warp parameters which structures took a lot of deformation or in this case information to map in which areas turned out to be the same place they were in the male. But this gives me an idea for another session, the gender politics of the neuroinformatics of neuroanatomy, but thank you very much, Mike, for a wonderful talk. Applause.