cg

changeset 33:6d023f15572e

.
author bshanks@bshanks.dyndns.org
date Mon Apr 13 14:53:12 2009 -0700 (16 years ago)
parents 70bdcd6c9972
children c435e5da5211
files grant.doc grant.html grant.odt grant.pdf grant.txt
line diff
1.1 Binary file grant.doc has changed
2.1 --- a/grant.html Mon Apr 13 14:31:11 2009 -0700 2.2 +++ b/grant.html Mon Apr 13 14:53:12 2009 -0700 2.3 @@ -1,25 +1,21 @@ 2.4 -Principal Investigator/Program Director(Last, First, Middle): Stevens, Charles F.___ 2.5 Specific aims 2.6 -Massive new datasets obtained with techniques such as in situ hybridization (ISH) and BAC-transgenics allow the expression 2.7 -levels of many genes at many locations to be compared. Our goal is to develop automated methods to relate spatial variation 2.8 -in gene expression to anatomy. We want to find marker genes for specific anatomical regions, and also to draw new anatomical 2.9 -maps based on gene expression patterns. We have three specific aims: 2.10 +Massive new datasets obtained with techniques such as in situ hybridization (ISH) and BAC-transgenics allow the expres- 2.11 +sion levels of many genes at many locations to be compared. Our goal is to develop automated methods to relate spatial 2.12 +variation in gene expression to anatomy. We want to find marker genes for specific anatomical regions, and also to draw 2.13 +new anatomical maps based on gene expression patterns. We have three specific aims: 2.14 (1) develop an algorithm to screen spatial gene expression data for combinations of marker genes which selectively target 2.15 anatomical regions 2.16 (2) develop an algorithm to suggest new ways of carving up a structure into anatomical subregions, based on spatial 2.17 patterns in gene expression 2.18 -(3) create a 2-D “flat map” dataset of the mouse cerebral cortex that contains a flattened version of the Allen Mouse Brain 2.19 -Atlas ISH data, as well as the boundaries of cortical anatomical areas. Use this dataset to validate the methods developed 2.20 -in (1) and (2). 2.21 +(3) create a 2-D “flat map” dataset of the mouse cerebral cortex that contains a flattened version of the Allen Mouse 2.22 +Brain Atlas ISH data, as well as the boundaries of cortical anatomical areas. Use this dataset to validate the methods 2.23 +developed in (1) and (2). 2.24 In addition to validating the usefulness of the algorithms, the application of these methods to cerebral cortex will produce 2.25 immediate benefits, because there are currently no known genetic markers for many cortical areas. The results of the project 2.26 -will support the development of new ways to selectively target cortical areas, and it will support the development of a method 2.27 -for identifying the cortical areal boundaries present in small tissue samples. 2.28 -All algorithms that we develop will be implemented in an open-source software toolkit. The toolkit, as well as the 2.29 +will support the development of new ways to selectively target cortical areas, and it will support the development of a 2.30 +method for identifying the cortical areal boundaries present in small tissue samples. 2.31 +All algorithms that we develop will be implemented in an open-source software toolkit. The toolkit, as well as the 2.32 machine-readable datasets developed in aim (3), will be published and freely available for others to use. 2.33 -_______________________________________________________________________________________________________ 2.34 -PHS 398/2590 (Rev. 09/04) Page 1 ___ Continuation Format Page 2.35 - Principal Investigator/Program Director(Last, First, Middle): Stevens, Charles F.___ 2.36 Background and significance 2.37 Aim 1 2.38 Machine learning terminology: supervised learning 2.39 @@ -30,21 +26,21 @@ 2.40 to which subregion each voxel within the structure belongs. We call this a classification task, because each voxel is being 2.41 assigned to a class (namely, its subregion). 2.42 Therefore, an understanding of the relationship between the combination of their expression levels and the locations of 2.43 -the subregions may be expressed as a function. The input to this function is a voxel, along with the gene expression levels 2.44 +the subregions may be expressed as a function. The input to this function is a voxel, along with the gene expression levels 2.45 within that voxel; the output is the subregional identity of the target voxel, that is, the subregion to which the target voxel 2.46 belongs. We call this function a classifier. In general, the input to a classifier is called an instance, and the output is called 2.47 a label (or a class label). 2.48 The object of aim 1 is not to produce a single classifier, but rather to develop an automated method for determining a 2.49 classifier for any known anatomical structure. Therefore, we seek a procedure by which a gene expression dataset may be 2.50 analyzed in concert with an anatomical atlas in order to produce a classifier. Such a procedure is a type of a machine learning 2.51 -procedure. The construction of the classifier is called training (also learning), and the initial gene expression dataset used in 2.52 -the construction of the classifier is called training data. 2.53 +procedure. The construction of the classifier is called training (also learning), and the initial gene expression dataset used 2.54 +in the construction of the classifier is called training data. 2.55 In the machine learning literature, this sort of procedure may be thought of as a supervised learning task, defined as a 2.56 task in which the goal is to learn a mapping from instances to labels, and the training data consists of a set of instances 2.57 (voxels) for which the labels (subregions) are known. 2.58 Each gene expression level is called a feature, and the selection of which genes1 to include is called feature selection. 2.59 -Feature selection is one component of the task of learning a classifier. Some methods for learning classifiers start out with a 2.60 -separate feature selection phase, whereas other methods combine feature selection with other aspects of training. 2.61 +Feature selection is one component of the task of learning a classifier. Some methods for learning classifiers start out with 2.62 +a separate feature selection phase, whereas other methods combine feature selection with other aspects of training. 2.63 One class of feature selection methods assigns some sort of score to each candidate gene. The top-ranked genes are then 2.64 chosen. Some scoring measures can assign a score to a set of selected genes, not just to a single gene; in this case, a dynamic 2.65 procedure may be used in which features are added and subtracted from the selected set depending on how much they raise 2.66 @@ -69,15 +65,12 @@ 2.67 in which it is not feasible to gather data about every gene. For example, if we want to use the expression of marker genes as 2.68 a trigger for some regionally-targeted intervention, then our intervention must contain a molecular mechanism to check the 2.69 expression level of each marker gene before it triggers. It is currently infeasible to design a molecular trigger that checks the 2.70 -_________________________________________ 2.71 - 1Strictly speaking, the features are gene expression levels, but we’ll call them genes. 2.72 -_______________________________________________________________________________________________________ 2.73 -PHS 398/2590 (Rev. 09/04) Page 2 ___ Continuation Format Page 2.74 - Principal Investigator/Program Director(Last, First, Middle): Stevens, Charles F.___ 2.75 -level of more than a handful of genes. Similarly, if the goal is to develop a procedure to do ISH on tissue samples in order 2.76 +level of more than a handful of genes. Similarly, if the goal is to develop a procedure to do ISH on tissue samples in order 2.77 to label their anatomy, then it is infeasible to label more than a few genes. Therefore, we must select only a few genes as 2.78 features. 2.79 Principle 3: Use geometry in feature selection 2.80 +_________________________________________ 2.81 + 1Strictly speaking, the features are gene expression levels, but we’ll call them genes. 2.82 When doing feature selection with score-based methods, the simplest thing to do would be to score the performance of 2.83 each voxel by itself and then combine these scores (pointwise scoring). A more powerful approach is to also use information 2.84 about the geometric relations between each voxel and its neighbors; this requires non-pointwise, local scoring methods. See 2.85 @@ -85,7 +78,8 @@ 2.86 Principle 4: Work in 2-D whenever possible 2.87 There are many anatomical structures which are commonly characterized in terms of a two-dimensional manifold. When 2.88 it is known that the structure that one is looking for is two-dimensional, the results may be improved by allowing the analysis 2.89 -algorithm to take advantage of this prior knowledge. In addition, it is easier for humans to visualize and work with 2-D data. 2.90 +algorithm to take advantage of this prior knowledge. In addition, it is easier for humans to visualize and work with 2-D 2.91 +data. 2.92 Therefore, when possible, the instances should represent pixels, not voxels. 2.93 Aim 2 2.94 Machine learning terminology: clustering 2.95 @@ -94,38 +88,35 @@ 2.96 together. A set of similar instances is called a cluster, and the activity of finding grouping the data into clusters is called 2.97 clustering or cluster analysis. 2.98 The task of deciding how to carve up a structure into anatomical subregions can be put into these terms. The instances 2.99 -are once again voxels (or pixels) along with their associated gene expression profiles. We make the assumption that voxels 2.100 -from the same subregion have similar gene expression profiles, at least compared to the other subregions. This means that 2.101 -clustering voxels is the same as finding potential subregions; we seek a partitioning of the voxels into subregions, that is, into 2.102 -clusters of voxels with similar gene expression. 2.103 +are once again voxels (or pixels) along with their associated gene expression profiles. We make the assumption that voxels 2.104 +from the same subregion have similar gene expression profiles, at least compared to the other subregions. This means that 2.105 +clustering voxels is the same as finding potential subregions; we seek a partitioning of the voxels into subregions, that is, 2.106 +into clusters of voxels with similar gene expression. 2.107 It is desirable to determine not just one set of subregions, but also how these subregions relate to each other, if at all; 2.108 -perhaps some of the subregions are more similar to each other than to the rest, suggesting that, although at a fine spatial scale 2.109 -they could be considered separate, on a coarser spatial scale they could be grouped together into one large subregion. This 2.110 -suggests the outcome of clustering may be a hierarchial tree of clusters, rather than a single set of clusters which partition 2.111 -the voxels. This is called hierarchial clustering. 2.112 +perhaps some of the subregions are more similar to each other than to the rest, suggesting that, although at a fine spatial 2.113 +scale they could be considered separate, on a coarser spatial scale they could be grouped together into one large subregion. 2.114 +This suggests the outcome of clustering may be a hierarchial tree of clusters, rather than a single set of clusters which 2.115 +partition the voxels. This is called hierarchial clustering. 2.116 Similarity scores 2.117 A crucial choice when designing a clustering method is how to measure similarity, across either pairs of instances, or 2.118 -clusters, or both. There is much overlap between scoring methods for feature selection (discussed above under Aim 1) and 2.119 +clusters, or both. There is much overlap between scoring methods for feature selection (discussed above under Aim 1) and 2.120 scoring methods for similarity. 2.121 Spatially contiguous clusters; image segmentation 2.122 -We have shown that aim 2 is a type of clustering task. In fact, it is a special type of clustering task because we have an 2.123 -additional constraint on clusters; voxels grouped together into a cluster must be spatially contiguous. In Preliminary Results, 2.124 -we show that one can get reasonable results without enforcing this constraint, however, we plan to compare these results 2.125 -against other methods which guarantee contiguous clusters. 2.126 +We have shown that aim 2 is a type of clustering task. In fact, it is a special type of clustering task because we have 2.127 +an additional constraint on clusters; voxels grouped together into a cluster must be spatially contiguous. In Preliminary 2.128 +Results, we show that one can get reasonable results without enforcing this constraint, however, we plan to compare these 2.129 +results against other methods which guarantee contiguous clusters. 2.130 Perhaps the biggest source of continguous clustering algorithms is the field of computer vision, which has produced a 2.131 -variety of image segmentation algorithms. Image segmentation is the task of partitioning the pixels in a digital image into 2.132 +variety of image segmentation algorithms. Image segmentation is the task of partitioning the pixels in a digital image into 2.133 clusters, usually contiguous clusters. Aim 2 is similar to an image segmentation task. There are two main differences; in 2.134 our task, there are thousands of color channels (one for each gene), rather than just three. There are imaging tasks which 2.135 -use more than three colors, however, for example multispectral imaging and hyperspectral imaging, which are often used to 2.136 -process satellite imagery. A more crucial difference is that there are various cues which are appropriate for detecting sharp 2.137 -object boundaries in a visual scene but which are not appropriate for segmenting abstract spatial data such as gene expression. 2.138 -Although many image segmentation algorithms can be expected to work well for segmenting other sorts of spatially arranged 2.139 -data, some of these algorithms are specialized for visual images. 2.140 +use more than three colors, however, for example multispectral imaging and hyperspectral imaging, which are often used 2.141 +to process satellite imagery. A more crucial difference is that there are various cues which are appropriate for detecting 2.142 +sharp object boundaries in a visual scene but which are not appropriate for segmenting abstract spatial data such as gene 2.143 +expression. Although many image segmentation algorithms can be expected to work well for segmenting other sorts of 2.144 +spatially arranged data, some of these algorithms are specialized for visual images. 2.145 Dimensionality reduction 2.146 -_______________________________________________________________________________________________________ 2.147 -PHS 398/2590 (Rev. 09/04) Page 3 ___ Continuation Format Page 2.148 - Principal Investigator/Program Director(Last, First, Middle): Stevens, Charles F.___ 2.149 -Unlike aim 1, there is no externally-imposed need to select only a handful of informative genes for inclusion in the 2.150 +Unlike aim 1, there is no externally-imposed need to select only a handful of informative genes for inclusion in the 2.151 instances. However, some clustering algorithms perform better on small numbers of features. There are techniques which 2.152 “summarize” a larger number of features using a smaller number of features; these techniques go by the name of feature 2.153 extraction or dimensionality reduction. The small set of features that such a technique yields is called the reduced feature 2.154 @@ -134,9 +125,9 @@ 2.155 feature set do not necessarily correspond to genes; each feature in the reduced set may be any function of the set of gene 2.156 expression levels. 2.157 Another use for dimensionality reduction is to visualize the relationships between subregions. For example, one might 2.158 -want to make a 2-D plot upon which each subregion is represented by a single point, and with the property that subregions 2.159 +want tomake a 2-D plot upon which each subregion is represented by a single point, and with the property that subregions 2.160 with similar gene expression profiles should be nearby on the plot (that is, the property that distance between pairs of points 2.161 -in the plot should be proportional to some measure of dissimilarity in gene expression). It is likely that no arrangement of 2.162 +in the plot should be proportional to some measure of dissimilarity in gene expression). It is likely that no arrangement of 2.163 the points on a 2-D plan will exactly satisfy this property – however, dimensionality reduction techniques allow one to find 2.164 arrangements of points that approximately satisfy that property. Note that in this application, dimensionality reduction 2.165 is being applied after clustering; whereas in the previous paragraph, we were talking about using dimensionality reduction 2.166 @@ -149,91 +140,86 @@ 2.167 Gene clusters could also be used to directly yield a clustering on instances. This is because many genes have an expression 2.168 pattern which seems to pick out a single, spatially continguous subregion. Therefore, it seems likely that an anatomically 2.169 interesting subregion will have multiple genes which each individually pick it out2. This suggests the following procedure: 2.170 -cluster together genes which pick out similar subregions, and then to use the more popular common subregions as the 2.171 -final clusters. In the Preliminary Data we show that a number of anatomically recognized cortical regions, as well as some 2.172 +cluster together genes which pick out similar subregions, and then to use the more popular common subregions as the 2.173 +final clusters. In the Preliminary Data we show that a number of anatomically recognized cortical regions, as well as some 2.174 “superregions” formed by lumping together a few regions, are associated with gene clusters in this fashion. 2.175 Aim 3 2.176 Background 2.177 -The cortex is divided into areas and layers. To a first approximation, the parcellation of the cortex into areas can be drawn 2.178 -as a 2-D map on the surface of the cortex. In the third dimension, the boundaries between the areas continue downwards 2.179 -into the cortical depth, perpendicular to the surface. The layer boundaries run parallel to the surface. One can picture an 2.180 -area of the cortex as a slice of many-layered cake. 2.181 +The cortex is divided into areas and layers. To a first approximation, the parcellation of the cortex into areas can 2.182 +be drawn as a 2-D map on the surface of the cortex. In the third dimension, the boundaries between the areas continue 2.183 +downwards into the cortical depth, perpendicular to the surface. The layer boundaries run parallel to the surface. One can 2.184 +picture an area of the cortex as a slice of many-layered cake. 2.185 Although it is known that different cortical areas have distinct roles in both normal functioning and in disease processes, 2.186 there are no known marker genes for many cortical areas. When it is necessary to divide a tissue sample into cortical areas, 2.187 this is a manual process that requires a skilled human to combine multiple visual cues and interpret them in the context of 2.188 their approximate location upon the cortical surface. 2.189 -Even the questions of how many areas should be recognized in cortex, and what their arrangement is, are still not 2.190 -completely settled. A proposed division of the cortex into areas is called a cortical map. In the rodent, the lack of a 2.191 -single agreed-upon map can be seen by contrasting the recent maps given by Swanson?? on the one hand, and Paxinos 2.192 -and Franklin?? on the other. While the maps are certainly very similar in their general arrangement, significant differences 2.193 +Even the questions of how many areas should be recognized in cortex, and what their arrangement is, are still not 2.194 +completely settled. A proposed division of the cortex into areas is called a cortical map. In the rodent, the lack of a 2.195 +single agreed-upon map can be seen by contrasting the recent maps given by Swanson[?] on the one hand, and Paxinos 2.196 +and Franklin[?] on the other. While the maps are certainly very similar in their general arrangement, significant differences 2.197 remain in the details. 2.198 Significance 2.199 The method developed in aim (1) will be applied to each cortical area to find a set of marker genes such that the 2.200 -combinatorial expression pattern of those genes uniquely picks out the target area. Finding marker genes will be useful for 2.201 +combinatorial expression pattern of those genes uniquely picks out the target area. Finding marker genes will be useful for 2.202 drug discovery as well as for experimentation because marker genes can be used to design interventions which selectively 2.203 target individual cortical areas. 2.204 -_______________ 2.205 - 2This would seem to contradict our finding in aim 1 that some cortical areas are combinatorially coded by multiple genes. However, it is possible 2.206 -that the currently accepted cortical maps divide the cortex into subregions which are unnatural from the point of view of gene expression; perhaps 2.207 -there is some other way to map the cortex for which each subregion can be identified by single genes. 2.208 -_______________________________________________________________________________________________________ 2.209 -PHS 398/2590 (Rev. 09/04) Page 4 ___ Continuation Format Page 2.210 - Principal Investigator/Program Director(Last, First, Middle): Stevens, Charles F.___ 2.211 The application of the marker gene finding algorithm to the cortex will also support the development of new neuroanatom- 2.212 -ical methods. In addition to finding markers for each individual cortical areas, we will find a small panel of genes that can 2.213 -find many of the areal boundaries at once. This panel of marker genes will allow the development of an ISH protocol that 2.214 +ical methods. In addition to finding markers for each individual cortical areas, we will find a small panel of genes that can 2.215 +find many of the areal boundaries at once. This panel of marker genes will allow the development of an ISH protocol that 2.216 will allow experimenters to more easily identify which anatomical areas are present in small samples of cortex. 2.217 -The method developed in aim (3) will provide a genoarchitectonic viewpoint that will contribute to the creation of a better 2.218 -map. The development of present-day cortical maps was driven by the application of histological stains. It is conceivable 2.219 -that if a different set of stains had been available which identified a different set of features, then the today’s cortical maps 2.220 -would have come out differently. Since the number of classes of stains is small compared to the number of genes, it is likely 2.221 -that there are many repeated, salient spatial patterns in the gene expression which have not yet been captured by any stain. 2.222 -Therefore, current ideas about cortical anatomy need to incorporate what we can learn from looking at the patterns of gene 2.223 -expression. 2.224 +The method developed in aim (3) will provide a genoarchitectonic viewpoint that will contribute to the creation of 2.225 +a better map. The development of present-day cortical maps was driven by the application of histological stains. It is 2.226 +conceivable that if a different set of stains had been available which identified a different set of features, then the today’s 2.227 +cortical maps would have come out differently. Since the number of classes of stains is small compared to the number of 2.228 +genes, it is likely that there are many repeated, salient spatial patterns in the gene expression which have not yet been 2.229 +captured by any stain. Therefore, current ideas about cortical anatomy need to incorporate what we can learn from looking 2.230 +at the patterns of gene expression. 2.231 While we do not here propose to analyze human gene expression data, it is conceivable that the methods we propose to 2.232 develop could be used to suggest modifications to the human cortical map as well. 2.233 +_________________________________________ 2.234 + 2This would seem to contradict our finding in aim 1 that some cortical areas are combinatorially coded by multiple genes. However, it is 2.235 +possible that the currently accepted cortical maps divide the cortex into subregions which are unnatural from the point of view of gene expression; 2.236 +perhaps there is some other way to map the cortex for which each subregion can be identified by single genes. 2.237 Related work 2.238 There does not appear to be much work on the automated analysis of spatial gene expression data. 2.239 There is a substantial body of work on the analysis of gene expression data, however, most of this concerns gene expression 2.240 data which is not fundamentally spatial. 2.241 As noted above, there has been much work on both supervised learning and clustering, and there are many available 2.242 algorithms for each. However, the completion of Aims 1 and 2 involves more than just choosing between a set of existing 2.243 -algorithms, and will constitute a substantial contribution to biology. The algorithms require the scientist to provide a 2.244 +algorithms, and will constitute a substantial contribution to biology. The algorithms require the scientist to provide a 2.245 framework for representing the problem domain, and the way that this framework is set up has a large impact on performance. 2.246 Creating a good framework can require creatively reconceptualizing the problem domain, and is not merely a mechanical 2.247 “fine-tuning” of numerical parameters. For example, we believe that domain-specific scoring measures (such as gradient 2.248 similarity, which is discussed in Preliminary Work) may be necessary in order to achieve the best results in this application. 2.249 We are aware of two existing efforts to relate spatial gene expression data to anatomy through computational methods. 2.250 -[? ] describes an analysis of the anatomy of the hippocampus using the ABA dataset. In addition to manual analysis, 2.251 +[3 ] describes an analysis of the anatomy of the hippocampus using the ABA dataset. In addition to manual analysis, 2.252 two clustering methods were employed, a modified Non-negative Matrix Factorization (NNMF), and a hierarchial recursive 2.253 bifurcation clustering scheme based on correlation as the similarity score. The paper yielded impressive results, proving the 2.254 usefulness of such research. We have run NNMF on the cortical dataset and while the results are promising (see Preliminary 2.255 Data), we think that it will be possible to find a better method3 (we also think that more automation of the parts that this 2.256 paper’s authors did manually will be possible). 2.257 -[? ] describes AGEA, ”Anatomic Gene Expression Atlas”. AGEA is an analysis tool for the ABA dataset. AGEA has 2.258 +[2 ] describes AGEA, ”Anatomic Gene Expression Atlas”. AGEA is an analysis tool for the ABA dataset. AGEA has 2.259 three components: 2.260 -* Gene Finder: The user selects a seed voxel and the system (1) chooses a cluster which includes the seed voxel, (2) yields 2.261 -a list of genes which are overexpressed in that cluster. 2.262 +* Gene Finder: The user selects a seed voxel and the system (1) chooses a cluster which includes the seed voxel, (2) 2.263 +yields a list of genes which are overexpressed in that cluster. 2.264 * Correlation: The user selects a seed voxel and the shows the user how much correlation there is between the gene 2.265 expression profile of the seed voxel and every other voxel. 2.266 -* Clusters: AGEA includes a precomputed hierarchial clustering of voxels based on a recursive bifurcation algorithm with 2.267 -correlation as the similarity metric. 2.268 +* Clusters: AGEA includes a precomputed hierarchial clustering of voxels based on a recursive bifurcation algorithm 2.269 +with correlation as the similarity metric. 2.270 At first glance AGEA seems similar to this proposal, but in fact it is different. 2.271 Gene Finder is different from our Aim 1 in at least four ways. First, although the user chooses a seed voxel, Gene Finder, 2.272 not the user, chooses the cluster for which genes will be found, and in our experience it never chooses cortical areas, instead 2.273 preferring cortical layers. Therefore, Gene Finder cannot be used to find marker genes for cortical areas. Second, Gene Finder 2.274 -finds only single genes, whereas we will also look for combinations of genes. Third, gene finder can only use overexpression 2.275 +finds only single genes, whereas we will also look for combinations of genes. Third, gene finder can only use overexpression 2.276 as a marker, whereas we will also look for underexpression. Fourth, Gene Finder uses a simple pointwise metric (“expression 2.277 energy ratio”, which captures overexpression), whereas we will also use geometric metrics such as gradient similarity. 2.278 The hierarchial clustering is different from our Aim 2 in at least two ways. todo 2.279 _________________________________________ 2.280 3We ran “vanilla” NNMF, whereas the paper under discussion used a modified method. Their main modification consisted of adding a soft 2.281 spatial contiguity constraint. However, on our dataset, NNMF naturally produced spatially contiguous clusters, so no additional constraint was 2.282 -needed. The paper under discussion mentions that they also tried a hierarchial variant of NNMF, but since they didn’t report its results, we assume 2.283 -that those result were not any more impressive than the results of the non-hierarchial variant. 2.284 -_______________________________________________________________________________________________________ 2.285 -PHS 398/2590 (Rev. 09/04) Page 5 ___ Continuation Format Page 2.286 - Principal Investigator/Program Director(Last, First, Middle): Stevens, Charles F.___ 2.287 +needed. The paper under discussion mentions that they also tried a hierarchial variant of NNMF, but since they didn’t report its results, we 2.288 +assume that those result were not any more impressive than the results of the non-hierarchial variant. 2.289 + 2.290 2.291 2.292 Figure 1: Upper left: wwc1. Upper right: mtif2. Lower left: wwc1 + mtif2 (each pixel’s value on the lower left is the sum 2.293 @@ -251,47 +237,48 @@ 2.294 natorially. according to logistic regression, gene wwc14 is the best fit single gene for predicting whether or not a pixel on 2.295 the cortical surface belongs to the motor area (area MO). The upper-left picture in Figure shows wwc1’s spatial expression 2.296 pattern over the cortex. The lower-right boundary of MO is represented reasonably well by this gene, however the gene 2.297 -overshoots the upper-left boundary. This flattened 2-D representation does not show it, but the area corresponding to the 2.298 +overshoots the upper-left boundary. This flattened 2-D representation does not show it, but the area corresponding to the 2.299 overshoot is the medial surface of the cortex. MO is only found on the lateral surface (todo). 2.300 Gnee mtif25 is shown in figure the upper-right of Fig. . Mtif2 captures MO’s upper-left boundary, but not its lower-right 2.301 -boundary. Mtif2 does not express very much on the medial surface. By adding together the values at each pixel in these two 2.302 -figures, we get the lower-left of Figure . This combination captures area MO much better than any single gene. 2.303 +boundary. Mtif2 does not express very much on the medial surface. By adding together the values at each pixel in these 2.304 +two figures, we get the lower-left of Figure . This combination captures area MO much better than any single gene. 2.305 Correlation todo 2.306 Conditional entropy todo 2.307 Gradient similarity todo 2.308 Geometric and pointwise scoring methods provide complementary information 2.309 To show that local geometry can provide useful information that cannot be detected via pointwise analyses, consider Fig. 2.310 -. The top row of Fig. displays the 3 genes which most match area AUD, according to a pointwise method6. The bottom row 2.311 +. The top row of Fig. displays the 3 genes which most match area AUD, according to a pointwise method6. The bottom 2.312 +row displays the 3 genes which most match AUD according to a method which considers local geometry7 The pointwise 2.313 +method in the top row identifies genes which express more strongly in AUD than outside of it; its weakness is that this 2.314 _________________________________________ 2.315 4“WW, C2 and coiled-coil domain containing 1”; EntrezGene ID 211652 2.316 5“mitochondrial translational initiation factor 2”; EntrezGene ID 76784 2.317 6For each gene, a logistic regression in which the response variable was whether or not a surface pixel was within area AUD, and the predictor 2.318 variable was the value of the expression of the gene underneath that pixel. The resulting scores were used to rank the genes in terms of how well 2.319 -_______________________________________________________________________________________________________ 2.320 -PHS 398/2590 (Rev. 09/04) Page 6 ___ Continuation Format Page 2.321 - Principal Investigator/Program Director(Last, First, Middle): Stevens, Charles F.___ 2.322 +they predict area AUD. 2.323 + 7For each gene the gradient similarity (see section ??) between (a) a map of the expression of each gene on the cortical surface and (b) the 2.324 +shape of area AUD, was calculated, and this was used to rank the genes. 2.325 + 2.326 2.327 2.328 -Figure 2: The top row shows the three genes which (individually) best predict area AUD, according to logistic regression. 2.329 +Figure 2: The top row shows the three genes which (individually) best predict area AUD, according to logistic regression. 2.330 The bottom row shows the three genes which (individually) best match area AUD, according to gradient similarity. From 2.331 left to right and top to bottom, the genes are Ssr1, Efcbp1, Aph1a, Ptk7, Aph1a again, and Lepr 2.332 -displays the 3 genes which most match AUD according to a method which considers local geometry7 The pointwise method 2.333 -in the top row identifies genes which express more strongly in AUD than outside of it; its weakness is that this includes many 2.334 -areas which don’t have a salient border matching the areal border. The geometric method identifies genes whose salient 2.335 -expression border seems to partially line up with the border of AUD; its weakness is that this includes genes which don’t 2.336 -express over the entire area. Genes which have high rankings using both pointwise and border criteria, such as Aph1a in the 2.337 -example, may be particularly good markers. None of these genes are, individually, a perfect marker for AUD; we deliberately 2.338 -chose a “difficult” area in order to better contrast pointwise with geometric methods. 2.339 +includes many areas which don’t have a salient border matching the areal border. The geometric method identifies genes 2.340 +whose salient expression border seems to partially line up with the border of AUD; its weakness is that this includes genes 2.341 +which don’t express over the entire area. Genes which have high rankings using both pointwise and border criteria, such as 2.342 +Aph1a in the example, may be particularly good markers. None of these genes are, individually, a perfect marker for AUD; 2.343 +we deliberately chose a “difficult” area in order to better contrast pointwise with geometric methods. 2.344 Areas which can be identified by single genes 2.345 todo 2.346 Specific to Aim 1 (and Aim 3) 2.347 Forward stepwise logistic regression todo 2.348 SVM on all genes at once 2.349 In order to see how well one can do when looking at all genes at once, we ran a support vector machine to classify cortical 2.350 -surface pixels based on their gene expression profiles. We achieved classification accuracy of about 81%8. As noted above, 2.351 +surface pixels based on their gene expression profiles. We achieved classification accuracy of about 81%8. As noted above, 2.352 however, a classifier that looks at all the genes at once isn’t practically useful. 2.353 The requirement to find combinations of only a small number of genes limits us from straightforwardly applying many 2.354 -of the most simple techniques from the field of supervised machine learning. In the parlance of machine learning, our task 2.355 +of the most simple techniques from the field of supervised machine learning. In the parlance of machine learning, our task 2.356 combines feature selection with supervised learning. 2.357 Decision trees 2.358 todo 2.359 @@ -299,21 +286,12 @@ 2.360 Raw dimensionality reduction results 2.361 todo 2.362 (might want to incld nnMF since mentioned above) 2.363 -_________________________________________ 2.364 -they predict area AUD. 2.365 - 7For each gene the gradient similarity (see section ??) between (a) a map of the expression of each gene on the cortical surface and (b) the 2.366 -shape of area AUD, was calculated, and this was used to rank the genes. 2.367 - 85-fold cross-validation. 2.368 -_______________________________________________________________________________________________________ 2.369 -PHS 398/2590 (Rev. 09/04) Page 7 ___ Continuation Format Page 2.370 - Principal Investigator/Program Director(Last, First, Middle): Stevens, Charles F.___ 2.371 Dimensionality reduction plus K-means or spectral clustering 2.372 Many areas are captured by clusters of genes 2.373 todo 2.374 todo 2.375 -_______________________________________________________________________________________________________ 2.376 -PHS 398/2590 (Rev. 09/04) Page 8 ___ Continuation Format Page 2.377 - Principal Investigator/Program Director(Last, First, Middle): Stevens, Charles F.___ 2.378 +_________________________________________ 2.379 + 85-fold cross-validation. 2.380 Research plan 2.381 todo amongst other things: 2.382 Develop algorithms that find genetic markers for anatomical regions 2.383 @@ -324,10 +302,10 @@ 2.384 3.Extend the procedure to handle difficult areas by using combinatorial coding: for areas that cannot be identified by any 2.385 single gene, identify them with a handful of genes. We will consider both (a) algorithms that incrementally/greedily 2.386 combine single gene markers into sets, such as forward stepwise regression and decision trees, and also (b) supervised 2.387 -learning techniques which use soft constraints to minimize the number of features, such as sparse support vector 2.388 +learning techniques which use soft constraints to minimize the number of features, such as sparse support vector 2.389 machines. 2.390 -4.Extend the procedure to handle difficult areas by combining or redrawing the boundaries: An area may be difficult to 2.391 -identify because the boundaries are misdrawn, or because it does not “really” exist as a single area, at least on the 2.392 +4.Extend the procedure to handle difficult areas by combining or redrawing the boundaries: An area may be difficult 2.393 +to identify because the boundaries are misdrawn, or because it does not “really” exist as a single area, at least on the 2.394 genetic level. We will develop extensions to our procedure which (a) detect when a difficult area could be fit if its 2.395 boundary were redrawn slightly, and (b) detect when a difficult area could be combined with adjacent areas to create 2.396 a larger area which can be fit. 2.397 @@ -345,26 +323,34 @@ 2.398 4.Explore clustering algorithms applied to genes: including gene shaving, TODO 2.399 5.Develop an algorithm to use dimensionality reduction and/or hierarchial clustering to create anatomical maps 2.400 6.Run this algorithm on the cortex: present a hierarchial, genoarchitectonic map of the cortex 2.401 -_____________________ 2.402 +Bibliography & References Cited 2.403 +[1]D C Van Essen, H A Drury, J Dickson, J Harwell, D Hanlon, and C H Anderson. An integrated software suite for surface- 2.404 +based analyses of cerebral cortex. Journal of the American Medical Informatics Association: JAMIA, 8(5):443–59, 2001. 2.405 +PMID: 11522765. 2.406 +[2]Lydia Ng, Amy Bernard, Chris Lau, Caroline C Overly, Hong-Wei Dong, Chihchau Kuan, Sayan Pathak, Susan M Sunkin, 2.407 +Chinh Dang, Jason W Bohland, Hemant Bokil, Partha P Mitra, Luis Puelles, John Hohmann, David J Anderson, Ed S 2.408 +Lein, Allan R Jones, and Michael Hawrylycz. An anatomic gene expression atlas of the adult mouse brain. Nat Neurosci, 2.409 +12(3):356–362, March 2009. 2.410 +[3]Carol L. Thompson, Sayan D. Pathak, Andreas Jeromin, Lydia L. Ng, Cameron R. MacPherson, Marty T. Mortrud, 2.411 +Allison Cusick, Zackery L. Riley, Susan M. Sunkin, Amy Bernard, Ralph B. Puchalski, Fred H. Gage, Allan R. Jones, 2.412 +Vladimir B. Bajic, Michael J. Hawrylycz, and Ed S. Lein. Genomic anatomy of the hippocampus. Neuron, 60(6):1010– 2.413 +1021, December 2008. 2.414 + 2.415 +_______________________________________________________________________________________________________ 2.416 stuff i dunno where to put yet (there is more scattered through grant-oldtext): 2.417 Principle 4: Work in 2-D whenever possible 2.418 - In anatomy, the manifold of interest is usually either defined by a combination of two relevant anatomical axes (todo), or 2.419 -by the surface of the structure (as is the case with the cortex). In the former case, the manifold of interest is a plane, but in 2.420 -the latter case it is curved. If the manifold is curved, there are various methods for mapping the manifold into a plane. 2.421 - The method that we will develop will begin by mapping the data into a 2-D plane. Although the manifold that charac- 2.422 -terized cortical areas is known to be the cortical surface, it remains to be seen which method of mapping the manifold into 2.423 -_______________________________________________________________________________________________________ 2.424 -PHS 398/2590 (Rev. 09/04) Page 9 ___ Continuation Format Page 2.425 - Principal Investigator/Program Director(Last, First, Middle): Stevens, Charles F.___ 2.426 -a plane is optimal for this application. We will compare mappings which attempt to preserve size (such as the one used by 2.427 -Caret?? ) with mappings which preserve angle (conformal maps). 2.428 -Although there is much 2-D organization in anatomy, there are also structures whose shape is fundamentally 3-dimensional. 2.429 + In anatomy, the manifold of interest is usually either defined by a combination of two relevant anatomical axes (todo), 2.430 +or by the surface of the structure (as is the case with the cortex). In the former case, the manifold of interest is a plane, but 2.431 +in the latter case it is curved. If the manifold is curved, there are various methods for mapping the manifold into a plane. 2.432 + The method that we will develop will begin by mapping the data into a 2-D plane. Although the manifold that 2.433 +characterized cortical areas is known to be the cortical surface, it remains to be seen which method of mapping the manifold 2.434 +into a plane is optimal for this application. We will compare mappings which attempt to preserve size (such as the one used 2.435 +by Caret[1]) with mappings which preserve angle (conformal maps). 2.436 + Although there is much 2-D organization in anatomy, there are also structures whose shape is fundamentally 3-dimensional. 2.437 If possible, we would like the method we develop to include a statistical test that warns the user if the assumption of 2-D 2.438 structure seems to be wrong. 2.439 -— 2.440 -note: 2.441 -do we need to cite: no known markers, impressive results? 2.442 -_______________________________________________________________________________________________________ 2.443 -PHS 398/2590 (Rev. 09/04) Page 10 ___ Continuation Format Page 2.444 - 2.445 - 2.446 + — 2.447 + note: 2.448 + do we need to cite: no known markers, impressive results? 2.449 + 2.450 +
3.1 Binary file grant.odt has changed
4.1 Binary file grant.pdf has changed
5.1 --- a/grant.txt Mon Apr 13 14:31:11 2009 -0700 5.2 +++ b/grant.txt Mon Apr 13 14:53:12 2009 -0700 5.3 @@ -1,5 +1,5 @@ 5.4 -\documentclass{nih} 5.5 -\piname{Stevens, Charles F.} 5.6 +\documentclass{nih-blank} 5.7 +%%\piname{Stevens, Charles F.} 5.8 5.9 == Specific aims == 5.10 5.11 @@ -119,7 +119,7 @@ 5.12 5.13 Although it is known that different cortical areas have distinct roles in both normal functioning and in disease processes, there are no known marker genes for many cortical areas. When it is necessary to divide a tissue sample into cortical areas, this is a manual process that requires a skilled human to combine multiple visual cues and interpret them in the context of their approximate location upon the cortical surface. 5.14 5.15 -Even the questions of how many areas should be recognized in cortex, and what their arrangement is, are still not completely settled. A proposed division of the cortex into areas is called a cortical map. In the rodent, the lack of a single agreed-upon map can be seen by contrasting the recent maps given by Swanson\ref{brain_swanson_2003} on the one hand, and Paxinos and Franklin\ref{mouse_paxinos_2001} on the other. While the maps are certainly very similar in their general arrangement, significant differences remain in the details. 5.16 +Even the questions of how many areas should be recognized in cortex, and what their arrangement is, are still not completely settled. A proposed division of the cortex into areas is called a cortical map. In the rodent, the lack of a single agreed-upon map can be seen by contrasting the recent maps given by Swanson\cite{brain_swanson_2003} on the one hand, and Paxinos and Franklin\cite{mouse_paxinos_2001} on the other. While the maps are certainly very similar in their general arrangement, significant differences remain in the details. 5.17 5.18 5.19 5.20 @@ -313,7 +313,12 @@ 5.21 5.22 5.23 5.24 - 5.25 +\newpage 5.26 + 5.27 +\bibliographystyle{plain} 5.28 +\bibliography{grant} 5.29 + 5.30 +\newpage 5.31 5.32 ---- 5.33 5.34 @@ -325,7 +330,7 @@ 5.35 5.36 In anatomy, the manifold of interest is usually either defined by a combination of two relevant anatomical axes (todo), or by the surface of the structure (as is the case with the cortex). In the former case, the manifold of interest is a plane, but in the latter case it is curved. If the manifold is curved, there are various methods for mapping the manifold into a plane. 5.37 5.38 -The method that we will develop will begin by mapping the data into a 2-D plane. Although the manifold that characterized cortical areas is known to be the cortical surface, it remains to be seen which method of mapping the manifold into a plane is optimal for this application. We will compare mappings which attempt to preserve size (such as the one used by Caret\ref{van_essen_integrated_2001}) with mappings which preserve angle (conformal maps). 5.39 +The method that we will develop will begin by mapping the data into a 2-D plane. Although the manifold that characterized cortical areas is known to be the cortical surface, it remains to be seen which method of mapping the manifold into a plane is optimal for this application. We will compare mappings which attempt to preserve size (such as the one used by Caret\cite{van_essen_integrated_2001}) with mappings which preserve angle (conformal maps). 5.40 5.41 Although there is much 2-D organization in anatomy, there are also structures whose shape is fundamentally 3-dimensional. If possible, we would like the method we develop to include a statistical test that warns the user if the assumption of 2-D structure seems to be wrong. 5.42 5.43 @@ -338,3 +343,6 @@ 5.44 note: 5.45 5.46 do we need to cite: no known markers, impressive results? 5.47 + 5.48 + 5.49 +