 It is my pleasure to introduce Dr. Douglas Bowden, Doug Bowden from the University of Washington. He's going to talk about Neuromaps, a multi-modality database in the shape of a brain. So Neuromaps is a multi-modality database in the shape of the brain. And in fact at this point we're talking about two brains, a macaque and the Waxholm mouse atlas. And I'd like to start off by going through the premises, our premises. And these are a little bit modified from Marianne's, but I think basically incompatible. The first is neuroanatomy provides the visual scaffold. And I'd say the common language for integrating information across scales, techniques, species, and so forth. Neuroinformatics has been building platforms for digital neuroanatomy for almost two decades. But up to now has not achieved the power of genomic informatics. And that is if you're a geneticist and you have established the nucleotide sequence of a new gene, you can go to any of a number of brain banks or gene banks, feed in that sequence, and you'll get back a list of all the brains of all the species that have that exact sequence. We should be able to do that in neuroanatomy, but at the present time we're not able to do that. Informatics requires that the information be made machine-computable and unambiguously, I would say human-computable or interpretable, we should say. The computational challenges of neuroinformatics are addressed, I would say, better in terms of information theory and communication theory than in terms of the truth-test computations of classical logic. And the main implication of that statement is that in order to have a comprehensive nomenclature, a comprehensive list of all the structures in the brain, one has to start from all the names that are used in neuroanatomy. And the ontology consists of relating those names to all of the structures, the subparts of the brain, that some legitimate neuroanatomist has defined. And the main component of that ontology is to specify as many of those relationships as you can. You need to be able to specify if a term is, if a new term is a synonym of one that already exists or if it's a homonym of one that already exists that's being applied to a different structure. And the concepts that are represented by these terms are really the definitions of the term. And you can, in this last paper, you saw it gets down to the point of defining the boundary between two structures right down to the cellular distribution level. So I'm saying that in contrast to classical logic, which works strictly with text, with words, word descriptions of structures, mathematical logic can be applied to your atlas and to your categories within your atlas. And it allows variance analyses that involve measurement and quantification, which is the hallmark of the scientific method. It allows, it gives statistical means of dealing with probabilistic truth. That is with the idea that as a scientist this is our best guess for today and it's held up for a hundred years and so we're believing it. But there's always a probability associated with that. And particularly when you get down to defining the lines between two structures, there's definitely probabilities you need to be able to deal with. You need an ontology that will allow you to deal with those. It allows parametric as well as non-parametric comparisons. And it allows particularly evaluation of the degree of overlap between spatial distributions. When the classical logic deals with categories and your smallest element that the classical logic deals with is one of the structures that you've seen that's defined with a boundary. But the basic element of a digital atlas is one point or one voxel or one pixel. And it defines structures by giving you a list of all the voxels that are in that structure. And that's a list that the human mind can't deal with. So we needed the computer to calculate the areas or the volumes or whatever. And the final statement I'd make on this is that one can compute anything by mathematical logic that one can compute by descriptive logic. But the opposite is not true. So I'm going to tell you about our system which is based on those premises. And the computational capability of a digital atlas depends on a fully segmented atlas and a comprehensive ontology of brain structure. And so what we have at this point is the Neuromaps digital brain atlas. The macaque brain atlas has been segmented down into approximately, I think, 400 structures. And we don't go to the, except in the hippocampus, don't really go to this layer with instructor or to the smallest sub nuclei. But it's defined down to those levels. And then we have the Wachsholm mouse brain atlas, which we are hoping, which has been subdivided, I think, into 50 or 70 structures. But we're hoping to import the segmentation of the Allen brain atlas so that one will be able to, which to my mind is the most comprehensive atlas that's available on the web at this point. And our ontology is in terms of Neuronames where we have defined the relationships between some 3,000 structures that are defined by various people. And we have standard names for those that each name is unique, which it's got to be if you're logic, whichever kind of logic you use, if it's going to be trustable. And some 16,000 synonyms in eight different languages. So a person can go into brain info with any legitimate name for a structure, and we will give you our standard name, and then you can explore the rest of the system and knowing that what you're looking for is this standard name. And we're now completing going through all of those definitions and determining which ones apply in the human, the macaque, the rat, and the mouse. So one will be able, using any logic, classical or mathematical, will be able to ask the question, okay, I'm interested in what genes are distributed in this particular area. The only data on that is probably going to be from the Allen Brain Institute atlas. But you're interested in the human. So you go into brain info, you look it up in the human, and then we'll tell you whether there is a mouse structure that is equivalent, as equivalent as anybody knows to this point, to that structure. So let's, I have a feeling my time is going fast, so let's just look at three questions here. What's required to use this system? What is required of a neuroscientist who wants to communicate data to others in a common format? Then what is required of neuroscientists who want to compare their data with other kinds of data from other sources, multimodality data? And then what is required of neuroinformaticists who want to create facilities that will fulfill the wants of these neuroscientists? So first, what's required of the scientist if he basically wants to put his data into a common format so other people can look at it? First, he conducts an experiment and creates an image or multiple images that shows a location of data relative to neuroanatomical features identifiable in a canonical digital brain atlas of the species. He downloads the canonical atlas and the mapping software from a neuroinformatics facility, downloads the tool, he maps the data to the atlas and the tool should provide the tools, sub-tools for doing that. Edits and labels the image of the map data into a format suitable for presentation or publication. And then downloads that back to his desktop and embeds that image in a slideshow and or a manuscript. And this is what neuromaps does. Neuromaps allows you to bring up the image of the MRI atlas in one screen and the image of your slide in another screen and you can go through the brain in coronal or whatever direction you want to go. You can tilt and rotate the atlas until you get the plane that most clearly matches your data. Then you identify equivalent landmarks in the two and then you can put in any number of landmarks. So if you're mainly interested in down in the hippocampus, you just put in landmarks, everything you can find that relates to that. Then you warp that image. What you actually do is you warp the atlas image to the experimental image and then you unwarp it back and that carries the data into your canonical atlas into the equivalent locations. Then you take that image and load it into the image editor and this is showing how we have mapped the cortical areas from one of the mouse atlases to the Waxholm brain. And now we can adjust the colors that we want to use for publication. We can add labels. We can determine whether the label is black on white or white on black and we can adjust the size of the labels. And you can adjust the dpi dots per inch and the size of the image for whatever journal, whatever their specifications are. And download this and then import it into your document. And this is the same sort of thing we've done with the monkey brain and we've mapped from the Paxonos macaque atlas. And this shows the kind of value added that you can provide to the investigator by using this system because if your system doesn't add value there are very few investigators who are going to bother to use it. And so in the upper left you see dopaminergic cells that have been mapped to the prefrontal cortex and if you look on the left side is the image that was published and on the right we have mapped those dots into the atlas superimposed the subdivisions of the atlas from the Paxonos atlas No, actually we subdivided on the basis of the sulcian gyri. But you can see that whereas in the first one, the left one it's pretty unclear where those cells are anatomically. In the one that after the mapping you can see that they're clustered right at the boundaries between the cerebral cortex and the white matter which is very interesting. Anyway, same sort of thing for a lesion study on the top right and on the bottom left I've plotted from some of our own work an electrode track where we were studying the... I think you get the point, let's go on to it. Okay, what's required of a neuroscientist who wants to compare map data with other kinds of data? I've separated this out because everything I've said so far is actually done on your own computer. But if you want to compare it with everybody else's you've got to upload it. So what you do is you map your data to the canonical digital brain atlas just as I described. You upload the images to the neuroinformatics facility together with identifying information such as a text, complete text description of the methods and metadata for unambiguous indexing of the data set. And the metadata terms for the metadata are pulled out of our neuro names. Then you request a comparison with all other data sets that are in the repository and a list of those that overlap your data by a certain proportion. And this can be done in 3D. Views overlap with other... Then if you go through the list and you find something that's particularly interesting you should be able to call up the image of that map, that map overlaid on yours so you can visually judge. And if that's still interesting you should be able to pull up the original image from which the mapping was done. So let's just go real quick to... This is what it looks like in neuromaps where we've overlaid Walker's area 46 and I think probably Broadman's corresponding area. If you want to know what the... And this is really the guts of the talk so I'm just going to go through it. I should say that we're at a very kind of critical point in our development of this system and that we're working on the strategic aims basically for the next five years. And if there's any part of what I'm going to say here that you think you have a tool that could do the job as well or better I'd like you to talk to me because maybe we can incorporate it into that project. So the informatics has to provide a canonical digital brain atlas of the species that people are going to be able to map into, needs to provide a comprehensive ontology and it's now available as neuronames, need to provide definitions and standard names that the ontology needs to provide definitions and standard names for every term. The facility needs to maintain a repository of all map data so that when somebody wants to know how much it overlaps with theirs it's already in the computer and it just creates the list. So with that, thank you very much. Set up if there's a question, we can have a question from the audience. I have one. Doug, does that put it out? Does neuromaps work in coordinate space or is it just its own internal space? Well, the neuromaps, does it put it out in what? Coordinates, I mean, are those structures referenced in the coordinates? Yes, the atlas has, as you're mapping in and as you've presented there are scales on the two sides. Does that map to stereotaxic space or is it just an internal? It would be very easy to define it in stereotaxic space. I mean, it's in the computer in stereotaxic space, yeah. Thank you. Okay, thank you, Doug.