| rev |
line source |
| bshanks@30 | 1 Principal Investigator/Program Director(Last, First, Middle): Stevens, Charles F.___
|
| bshanks@0 | 2 Specific aims
|
| bshanks@30 | 3 Massive new datasets obtained with techniques such as in situ hybridization (ISH) and BAC-transgenics allow the expression
|
| bshanks@30 | 4 levels of many genes at many locations to be compared. Our goal is to develop automated methods to relate spatial variation
|
| bshanks@30 | 5 in gene expression to anatomy. We want to find marker genes for specific anatomical regions, and also to draw new anatomical
|
| bshanks@30 | 6 maps based on gene expression patterns. We have three specific aims:
|
| bshanks@30 | 7 (1) develop an algorithm to screen spatial gene expression data for combinations of marker genes which selectively target
|
| bshanks@30 | 8 anatomical regions
|
| bshanks@30 | 9 (2) develop an algorithm to suggest new ways of carving up a structure into anatomical subregions, based on spatial
|
| bshanks@30 | 10 patterns in gene expression
|
| bshanks@30 | 11 (3) create a 2-D “flat map” dataset of the mouse cerebral cortex that contains a flattened version of the Allen Mouse Brain
|
| bshanks@30 | 12 Atlas ISH data, as well as the boundaries of cortical anatomical areas. Use this dataset to validate the methods developed
|
| bshanks@30 | 13 in (1) and (2).
|
| bshanks@30 | 14 In addition to validating the usefulness of the algorithms, the application of these methods to cerebral cortex will produce
|
| bshanks@30 | 15 immediate benefits, because there are currently no known genetic markers for many cortical areas. The results of the project
|
| bshanks@30 | 16 will support the development of new ways to selectively target cortical areas, and it will support the development of a method
|
| bshanks@30 | 17 for identifying the cortical areal boundaries present in small tissue samples.
|
| bshanks@30 | 18 All algorithms that we develop will be implemented in an open-source software toolkit. The toolkit, as well as the
|
| bshanks@30 | 19 machine-readable datasets developed in aim (3), will be published and freely available for others to use.
|
| bshanks@30 | 20 _______________________________________________________________________________________________________
|
| bshanks@30 | 21 PHS 398/2590 (Rev. 09/04) Page 1 ___ Continuation Format Page
|
| bshanks@30 | 22 Principal Investigator/Program Director(Last, First, Middle): Stevens, Charles F.___
|
| bshanks@30 | 23 Background and significance
|
| bshanks@30 | 24 Aim 1
|
| bshanks@30 | 25 Machine learning terminology: supervised learning
|
| bshanks@30 | 26 The task of looking for marker genes for anatomical subregions means that one is looking for a set of genes such that, if
|
| bshanks@30 | 27 the expression level of those genes is known, then the locations of the subregions can be inferred.
|
| bshanks@30 | 28 If we define the subregions so that they cover the entire anatomical structure to be divided, then instead of saying that we
|
| bshanks@30 | 29 are using gene expression to find the locations of the subregions, we may say that we are using gene expression to determine
|
| bshanks@30 | 30 to which subregion each voxel within the structure belongs. We call this a classification task, because each voxel is being
|
| bshanks@30 | 31 assigned to a class (namely, its subregion).
|
| bshanks@30 | 32 Therefore, an understanding of the relationship between the combination of their expression levels and the locations of
|
| bshanks@30 | 33 the subregions may be expressed as a function. The input to this function is a voxel, along with the gene expression levels
|
| bshanks@30 | 34 within that voxel; the output is the subregional identity of the target voxel, that is, the subregion to which the target voxel
|
| bshanks@30 | 35 belongs. We call this function a classifier. In general, the input to a classifier is called an instance, and the output is called
|
| bshanks@30 | 36 a label (or a class label).
|
| bshanks@30 | 37 The object of aim 1 is not to produce a single classifier, but rather to develop an automated method for determining a
|
| bshanks@30 | 38 classifier for any known anatomical structure. Therefore, we seek a procedure by which a gene expression dataset may be
|
| bshanks@30 | 39 analyzed in concert with an anatomical atlas in order to produce a classifier. Such a procedure is a type of a machine learning
|
| bshanks@30 | 40 procedure. The construction of the classifier is called training (also learning), and the initial gene expression dataset used in
|
| bshanks@30 | 41 the construction of the classifier is called training data.
|
| bshanks@30 | 42 In the machine learning literature, this sort of procedure may be thought of as a supervised learning task, defined as a
|
| bshanks@30 | 43 task in which the goal is to learn a mapping from instances to labels, and the training data consists of a set of instances
|
| bshanks@30 | 44 (voxels) for which the labels (subregions) are known.
|
| bshanks@30 | 45 Each gene expression level is called a feature, and the selection of which genes1 to include is called feature selection.
|
| bshanks@30 | 46 Feature selection is one component of the task of learning a classifier. Some methods for learning classifiers start out with a
|
| bshanks@30 | 47 separate feature selection phase, whereas other methods combine feature selection with other aspects of training.
|
| bshanks@30 | 48 One class of feature selection methods assigns some sort of score to each candidate gene. The top-ranked genes are then
|
| bshanks@30 | 49 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
|
| bshanks@30 | 50 procedure may be used in which features are added and subtracted from the selected set depending on how much they raise
|
| bshanks@30 | 51 the score. Such procedures are called “stepwise” or “greedy”.
|
| bshanks@30 | 52 Although the classifier itself may only look at the gene expression data within each voxel before classifying that voxel, the
|
| bshanks@30 | 53 learning algorithm which constructs the classifier may look over the entire dataset. We can categorize score-based feature
|
| bshanks@30 | 54 selection methods depending on how the score of calculated. Often the score calculation consists of assigning a sub-score to
|
| bshanks@30 | 55 each voxel, and then aggregating these sub-scores into a final score (the aggregation is often a sum or a sum of squares). If
|
| bshanks@30 | 56 only information from nearby voxels is used to calculate a voxel’s sub-score, then we say it is a local scoring method. If only
|
| bshanks@30 | 57 information from the voxel itself is used to calculate a voxel’s sub-score, then we say it is a pointwise scoring method.
|
| bshanks@30 | 58 Key questions when choosing a learning method are: What are the instances? What are the features? How are the
|
| bshanks@30 | 59 features chosen? Here are four principles that outline our answers to these questions.
|
| bshanks@30 | 60 Principle 1: Combinatorial gene expression It is too much to hope that every anatomical region of interest will be
|
| bshanks@30 | 61 identified by a single gene. For example, in the cortex, there are some areas which are not clearly delineated by any gene
|
| bshanks@30 | 62 included in the Allen Brain Atlas (ABA) dataset. However, at least some of these areas can be delineated by looking at
|
| bshanks@30 | 63 combinations of genes (an example of an area for which multiple genes are necessary and sufficient is provided in Preliminary
|
| bshanks@30 | 64 Results). Therefore, each instance should contain multiple features (genes).
|
| bshanks@30 | 65 Principle 2: Only look at combinations of small numbers of genes When the classifier classifies a voxel, it is
|
| bshanks@30 | 66 only allowed to look at the expression of the genes which have been selected as features. The more data that is available to
|
| bshanks@30 | 67 a classifier, the better that it can do. For example, perhaps there are weak correlations over many genes that add up to a
|
| bshanks@30 | 68 strong signal. So, why not include every gene as a feature? The reason is that we wish to employ the classifier in situations
|
| bshanks@30 | 69 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
|
| bshanks@30 | 70 a trigger for some regionally-targeted intervention, then our intervention must contain a molecular mechanism to check the
|
| bshanks@30 | 71 expression level of each marker gene before it triggers. It is currently infeasible to design a molecular trigger that checks the
|
| bshanks@30 | 72 _________________________________________
|
| bshanks@30 | 73 1Strictly speaking, the features are gene expression levels, but we’ll call them genes.
|
| bshanks@30 | 74 _______________________________________________________________________________________________________
|
| bshanks@30 | 75 PHS 398/2590 (Rev. 09/04) Page 2 ___ Continuation Format Page
|
| bshanks@30 | 76 Principal Investigator/Program Director(Last, First, Middle): Stevens, Charles F.___
|
| bshanks@30 | 77 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
|
| bshanks@30 | 78 to label their anatomy, then it is infeasible to label more than a few genes. Therefore, we must select only a few genes as
|
| bshanks@30 | 79 features.
|
| bshanks@30 | 80 Principle 3: Use geometry in feature selection
|
| bshanks@30 | 81 When doing feature selection with score-based methods, the simplest thing to do would be to score the performance of
|
| bshanks@30 | 82 each voxel by itself and then combine these scores (pointwise scoring). A more powerful approach is to also use information
|
| bshanks@30 | 83 about the geometric relations between each voxel and its neighbors; this requires non-pointwise, local scoring methods. See
|
| bshanks@30 | 84 Preliminary Results for evidence of the complementary nature of pointwise and local scoring methods.
|
| bshanks@30 | 85 Principle 4: Work in 2-D whenever possible
|
| bshanks@30 | 86 There are many anatomical structures which are commonly characterized in terms of a two-dimensional manifold. When
|
| bshanks@30 | 87 it is known that the structure that one is looking for is two-dimensional, the results may be improved by allowing the analysis
|
| bshanks@30 | 88 algorithm to take advantage of this prior knowledge. In addition, it is easier for humans to visualize and work with 2-D data.
|
| bshanks@30 | 89 Therefore, when possible, the instances should represent pixels, not voxels.
|
| bshanks@30 | 90 Aim 2
|
| bshanks@30 | 91 Machine learning terminology: clustering
|
| bshanks@30 | 92 If one is given a dataset consisting merely of instances, with no class labels, then analysis of the dataset is referred to as
|
| bshanks@30 | 93 unsupervised learning in the jargon of machine learning. One thing that you can do with such a dataset is to group instances
|
| bshanks@30 | 94 together. A set of similar instances is called a cluster, and the activity of finding grouping the data into clusters is called
|
| bshanks@30 | 95 clustering or cluster analysis.
|
| bshanks@30 | 96 The task of deciding how to carve up a structure into anatomical subregions can be put into these terms. The instances
|
| bshanks@30 | 97 are once again voxels (or pixels) along with their associated gene expression profiles. We make the assumption that voxels
|
| bshanks@30 | 98 from the same subregion have similar gene expression profiles, at least compared to the other subregions. This means that
|
| bshanks@30 | 99 clustering voxels is the same as finding potential subregions; we seek a partitioning of the voxels into subregions, that is, into
|
| bshanks@30 | 100 clusters of voxels with similar gene expression.
|
| bshanks@30 | 101 It is desirable to determine not just one set of subregions, but also how these subregions relate to each other, if at all;
|
| bshanks@30 | 102 perhaps some of the subregions are more similar to each other than to the rest, suggesting that, although at a fine spatial scale
|
| bshanks@30 | 103 they could be considered separate, on a coarser spatial scale they could be grouped together into one large subregion. This
|
| bshanks@30 | 104 suggests the outcome of clustering may be a hierarchial tree of clusters, rather than a single set of clusters which partition
|
| bshanks@30 | 105 the voxels. This is called hierarchial clustering.
|
| bshanks@30 | 106 Similarity scores
|
| bshanks@30 | 107 A crucial choice when designing a clustering method is how to measure similarity, across either pairs of instances, or
|
| bshanks@30 | 108 clusters, or both. There is much overlap between scoring methods for feature selection (discussed above under Aim 1) and
|
| bshanks@30 | 109 scoring methods for similarity.
|
| bshanks@30 | 110 Spatially contiguous clusters; image segmentation
|
| bshanks@30 | 111 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
|
| bshanks@30 | 112 additional constraint on clusters; voxels grouped together into a cluster must be spatially contiguous. In Preliminary Results,
|
| bshanks@30 | 113 we show that one can get reasonable results without enforcing this constraint, however, we plan to compare these results
|
| bshanks@30 | 114 against other methods which guarantee contiguous clusters.
|
| bshanks@30 | 115 Perhaps the biggest source of continguous clustering algorithms is the field of computer vision, which has produced a
|
| bshanks@30 | 116 variety of image segmentation algorithms. Image segmentation is the task of partitioning the pixels in a digital image into
|
| bshanks@30 | 117 clusters, usually contiguous clusters. Aim 2 is similar to an image segmentation task. There are two main differences; in
|
| bshanks@30 | 118 our task, there are thousands of color channels (one for each gene), rather than just three. There are imaging tasks which
|
| bshanks@30 | 119 use more than three colors, however, for example multispectral imaging and hyperspectral imaging, which are often used to
|
| bshanks@30 | 120 process satellite imagery. A more crucial difference is that there are various cues which are appropriate for detecting sharp
|
| bshanks@30 | 121 object boundaries in a visual scene but which are not appropriate for segmenting abstract spatial data such as gene expression.
|
| bshanks@30 | 122 Although many image segmentation algorithms can be expected to work well for segmenting other sorts of spatially arranged
|
| bshanks@30 | 123 data, some of these algorithms are specialized for visual images.
|
| bshanks@30 | 124 Dimensionality reduction
|
| bshanks@30 | 125 _______________________________________________________________________________________________________
|
| bshanks@30 | 126 PHS 398/2590 (Rev. 09/04) Page 3 ___ Continuation Format Page
|
| bshanks@30 | 127 Principal Investigator/Program Director(Last, First, Middle): Stevens, Charles F.___
|
| bshanks@30 | 128 Unlike aim 1, there is no externally-imposed need to select only a handful of informative genes for inclusion in the
|
| bshanks@30 | 129 instances. However, some clustering algorithms perform better on small numbers of features. There are techniques which
|
| bshanks@30 | 130 “summarize” a larger number of features using a smaller number of features; these techniques go by the name of feature
|
| bshanks@30 | 131 extraction or dimensionality reduction. The small set of features that such a technique yields is called the reduced feature
|
| bshanks@30 | 132 set. After the reduced feature set is created, the instances may be replaced by reduced instances, which have as their features
|
| bshanks@30 | 133 the reduced feature set rather than the original feature set of all gene expression levels. Note that the features in the reduced
|
| bshanks@30 | 134 feature set do not necessarily correspond to genes; each feature in the reduced set may be any function of the set of gene
|
| bshanks@30 | 135 expression levels.
|
| bshanks@30 | 136 Another use for dimensionality reduction is to visualize the relationships between subregions. For example, one might
|
| bshanks@30 | 137 want to make a 2-D plot upon which each subregion is represented by a single point, and with the property that subregions
|
| bshanks@30 | 138 with similar gene expression profiles should be nearby on the plot (that is, the property that distance between pairs of points
|
| bshanks@30 | 139 in the plot should be proportional to some measure of dissimilarity in gene expression). It is likely that no arrangement of
|
| bshanks@30 | 140 the points on a 2-D plan will exactly satisfy this property – however, dimensionality reduction techniques allow one to find
|
| bshanks@30 | 141 arrangements of points that approximately satisfy that property. Note that in this application, dimensionality reduction
|
| bshanks@30 | 142 is being applied after clustering; whereas in the previous paragraph, we were talking about using dimensionality reduction
|
| bshanks@30 | 143 before clustering.
|
| bshanks@30 | 144 Clustering genes rather than voxels
|
| bshanks@30 | 145 Although the ultimate goal is to cluster the instances (voxels or pixels), one strategy to achieve this goal is to first cluster
|
| bshanks@30 | 146 the features (genes). There are two ways that clusters of genes could be used.
|
| bshanks@30 | 147 Gene clusters could be used as part of dimensionality reduction: rather than have one feature for each gene, we could
|
| bshanks@30 | 148 have one reduced feature for each gene cluster.
|
| bshanks@30 | 149 Gene clusters could also be used to directly yield a clustering on instances. This is because many genes have an expression
|
| bshanks@30 | 150 pattern which seems to pick out a single, spatially continguous subregion. Therefore, it seems likely that an anatomically
|
| bshanks@30 | 151 interesting subregion will have multiple genes which each individually pick it out2. This suggests the following procedure:
|
| bshanks@30 | 152 cluster together genes which pick out similar subregions, and then to use the more popular common subregions as the
|
| bshanks@30 | 153 final clusters. In the Preliminary Data we show that a number of anatomically recognized cortical regions, as well as some
|
| bshanks@30 | 154 “superregions” formed by lumping together a few regions, are associated with gene clusters in this fashion.
|
| bshanks@30 | 155 Aim 3
|
| bshanks@30 | 156 Background
|
| bshanks@30 | 157 The cortex is divided into areas and layers. To a first approximation, the parcellation of the cortex into areas can be drawn
|
| bshanks@30 | 158 as a 2-D map on the surface of the cortex. In the third dimension, the boundaries between the areas continue downwards
|
| bshanks@30 | 159 into the cortical depth, perpendicular to the surface. The layer boundaries run parallel to the surface. One can picture an
|
| bshanks@30 | 160 area of the cortex as a slice of many-layered cake.
|
| bshanks@30 | 161 Although it is known that different cortical areas have distinct roles in both normal functioning and in disease processes,
|
| bshanks@30 | 162 there are no known marker genes for many cortical areas. When it is necessary to divide a tissue sample into cortical areas,
|
| bshanks@30 | 163 this is a manual process that requires a skilled human to combine multiple visual cues and interpret them in the context of
|
| bshanks@30 | 164 their approximate location upon the cortical surface.
|
| bshanks@30 | 165 Even the questions of how many areas should be recognized in cortex, and what their arrangement is, are still not
|
| bshanks@30 | 166 completely settled. A proposed division of the cortex into areas is called a cortical map. In the rodent, the lack of a
|
| bshanks@30 | 167 single agreed-upon map can be seen by contrasting the recent maps given by Swanson?? on the one hand, and Paxinos
|
| bshanks@30 | 168 and Franklin?? on the other. While the maps are certainly very similar in their general arrangement, significant differences
|
| bshanks@30 | 169 remain in the details.
|
| bshanks@30 | 170 Significance
|
| bshanks@30 | 171 The method developed in aim (1) will be applied to each cortical area to find a set of marker genes such that the
|
| bshanks@30 | 172 combinatorial expression pattern of those genes uniquely picks out the target area. Finding marker genes will be useful for
|
| bshanks@30 | 173 drug discovery as well as for experimentation because marker genes can be used to design interventions which selectively
|
| bshanks@30 | 174 target individual cortical areas.
|
| bshanks@30 | 175 _______________
|
| bshanks@30 | 176 2This would seem to contradict our finding in aim 1 that some cortical areas are combinatorially coded by multiple genes. However, it is possible
|
| bshanks@30 | 177 that the currently accepted cortical maps divide the cortex into subregions which are unnatural from the point of view of gene expression; perhaps
|
| bshanks@30 | 178 there is some other way to map the cortex for which each subregion can be identified by single genes.
|
| bshanks@30 | 179 _______________________________________________________________________________________________________
|
| bshanks@30 | 180 PHS 398/2590 (Rev. 09/04) Page 4 ___ Continuation Format Page
|
| bshanks@30 | 181 Principal Investigator/Program Director(Last, First, Middle): Stevens, Charles F.___
|
| bshanks@30 | 182 The application of the marker gene finding algorithm to the cortex will also support the development of new neuroanatom-
|
| bshanks@30 | 183 ical methods. In addition to finding markers for each individual cortical areas, we will find a small panel of genes that can
|
| bshanks@30 | 184 find many of the areal boundaries at once. This panel of marker genes will allow the development of an ISH protocol that
|
| bshanks@30 | 185 will allow experimenters to more easily identify which anatomical areas are present in small samples of cortex.
|
| bshanks@30 | 186 The method developed in aim (3) will provide a genoarchitectonic viewpoint that will contribute to the creation of a better
|
| bshanks@30 | 187 map. The development of present-day cortical maps was driven by the application of histological stains. It is conceivable
|
| bshanks@30 | 188 that if a different set of stains had been available which identified a different set of features, then the today’s cortical maps
|
| bshanks@30 | 189 would have come out differently. Since the number of classes of stains is small compared to the number of genes, it is likely
|
| bshanks@30 | 190 that there are many repeated, salient spatial patterns in the gene expression which have not yet been captured by any stain.
|
| bshanks@30 | 191 Therefore, current ideas about cortical anatomy need to incorporate what we can learn from looking at the patterns of gene
|
| bshanks@30 | 192 expression.
|
| bshanks@30 | 193 While we do not here propose to analyze human gene expression data, it is conceivable that the methods we propose to
|
| bshanks@30 | 194 develop could be used to suggest modifications to the human cortical map as well.
|
| bshanks@30 | 195 Related work
|
| bshanks@30 | 196 There does not appear to be much work on the automated analysis of spatial gene expression data.
|
| bshanks@30 | 197 There is a substantial body of work on the analysis of gene expression data, however, most of this concerns gene expression
|
| bshanks@30 | 198 data which is not fundamentally spatial.
|
| bshanks@30 | 199 As noted above, there has been much work on both supervised learning and clustering, and there are many available
|
| bshanks@30 | 200 algorithms for each. However, the completion of Aims 1 and 2 involves more than just choosing between a set of existing
|
| bshanks@30 | 201 algorithms, and will constitute a substantial contribution to biology. The algorithms require the scientist to provide a
|
| bshanks@30 | 202 framework for representing the problem domain, and the way that this framework is set up has a large impact on performance.
|
| bshanks@30 | 203 Creating a good framework can require creatively reconceptualizing the problem domain, and is not merely a mechanical
|
| bshanks@30 | 204 “fine-tuning” of numerical parameters. For example, we believe that domain-specific scoring measures (such as gradient
|
| bshanks@30 | 205 similarity, which is discussed in Preliminary Work) may be necessary in order to achieve the best results in this application.
|
| bshanks@30 | 206 We are aware of two existing efforts to relate spatial gene expression data to anatomy through computational methods.
|
| bshanks@30 | 207 [? ] describes an analysis of the anatomy of the hippocampus using the ABA dataset. In addition to manual analysis,
|
| bshanks@30 | 208 two clustering methods were employed, a modified Non-negative Matrix Factorization (NNMF), and a hierarchial bifurcation
|
| bshanks@30 | 209 clustering scheme based on correlation as the similarity score. The paper yielded impressive results, proving the usefulness
|
| bshanks@30 | 210 of such research. We have run NNMF on the cortical dataset and while the results are promising (see Preliminary Data), we
|
| bshanks@30 | 211 think that it will be possible to find a better method3 (we also think that more automation of the parts that this paper’s
|
| bshanks@30 | 212 authors did manually will be possible).
|
| bshanks@30 | 213 and [?] describes AGEA. todo
|
| bshanks@30 | 214 _____________
|
| bshanks@30 | 215 3We ran “vanilla” NNMF, whereas the paper under discussion used a modified method. Their main modification consisted of adding a soft
|
| bshanks@30 | 216 spatial contiguity constraint. However, on our dataset, NNMF naturally produced spatially contiguous clusters, so no additional constraint was
|
| bshanks@30 | 217 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
|
| bshanks@30 | 218 that those result were not any more impressive than the results of the non-hierarchial variant.
|
| bshanks@30 | 219 _______________________________________________________________________________________________________
|
| bshanks@30 | 220 PHS 398/2590 (Rev. 09/04) Page 5 ___ Continuation Format Page
|
| bshanks@30 | 221 Principal Investigator/Program Director(Last, First, Middle): Stevens, Charles F.___
|
| bshanks@30 | 222
|
| bshanks@26 | 223
|
| bshanks@30 | 224 Figure 1: Upper left: wwc1. Upper right: mtif2. Lower left: wwc1 + mtif2 (each pixel’s value on the lower left is the sum
|
| bshanks@30 | 225 of the corresponding pixels in the upper row). Within each picture, the vertical axis roughly corresponds to anterior at the
|
| bshanks@30 | 226 top and posterior at the bottom, and the horizontal axis roughly corresponds to medial at the left and lateral at the right.
|
| bshanks@30 | 227 The red outline is the boundary of region MO. Pixels are colored approximately according to the density of expressing cells
|
| bshanks@30 | 228 underneath each pixel, with red meaning a lot of expression and blue meaning little.
|
| bshanks@30 | 229 Preliminary work
|
| bshanks@30 | 230 Format conversion between SEV, MATLAB, NIFTI
|
| bshanks@30 | 231 todo
|
| bshanks@30 | 232 Flatmap of cortex
|
| bshanks@30 | 233 todo
|
| bshanks@30 | 234 Using combinations of multiple genes is necessary and sufficient to delineate some cortical areas
|
| bshanks@30 | 235 Here we give an example of a cortical area which is not marked by any single gene, but which can be identified combi-
|
| bshanks@30 | 236 natorially. according to logistic regression, gene wwc14 is the best fit single gene for predicting whether or not a pixel on
|
| bshanks@30 | 237 the cortical surface belongs to the motor area (area MO). The upper-left picture in Figure shows wwc1’s spatial expression
|
| bshanks@30 | 238 pattern over the cortex. The lower-right boundary of MO is represented reasonably well by this gene, however the gene
|
| bshanks@30 | 239 overshoots the upper-left boundary. This flattened 2-D representation does not show it, but the area corresponding to the
|
| bshanks@30 | 240 overshoot is the medial surface of the cortex. MO is only found on the lateral surface (todo).
|
| bshanks@30 | 241 Gnee mtif25 is shown in figure the upper-right of Fig. . Mtif2 captures MO’s upper-left boundary, but not its lower-right
|
| bshanks@30 | 242 boundary. Mtif2 does not express very much on the medial surface. By adding together the values at each pixel in these two
|
| bshanks@30 | 243 figures, we get the lower-left of Figure . This combination captures area MO much better than any single gene.
|
| bshanks@30 | 244 Correlation todo
|
| bshanks@30 | 245 Conditional entropy todo
|
| bshanks@30 | 246 Gradient similarity todo
|
| bshanks@30 | 247 Geometric and pointwise scoring methods provide complementary information
|
| bshanks@30 | 248 To show that local geometry can provide useful information that cannot be detected via pointwise analyses, consider Fig.
|
| bshanks@30 | 249 . The top row of Fig. displays the 3 genes which most match area AUD, according to a pointwise method6. The bottom row
|
| bshanks@30 | 250 _________________________________________
|
| bshanks@29 | 251 4“WW, C2 and coiled-coil domain containing 1”; EntrezGene ID 211652
|
| bshanks@29 | 252 5“mitochondrial translational initiation factor 2”; EntrezGene ID 76784
|
| bshanks@30 | 253 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
|
| bshanks@30 | 254 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
|
| bshanks@30 | 255 _______________________________________________________________________________________________________
|
| bshanks@30 | 256 PHS 398/2590 (Rev. 09/04) Page 6 ___ Continuation Format Page
|
| bshanks@30 | 257 Principal Investigator/Program Director(Last, First, Middle): Stevens, Charles F.___
|
| bshanks@30 | 258
|
| bshanks@30 | 259
|
| bshanks@30 | 260 Figure 2: The top row shows the three genes which (individually) best predict area AUD, according to logistic regression.
|
| bshanks@30 | 261 The bottom row shows the three genes which (individually) best match area AUD, according to gradient similarity. From
|
| bshanks@30 | 262 left to right and top to bottom, the genes are Ssr1, Efcbp1, Aph1a, Ptk7, Aph1a again, and Lepr
|
| bshanks@30 | 263 displays the 3 genes which most match AUD according to a method which considers local geometry7 The pointwise method
|
| bshanks@30 | 264 in the top row identifies genes which express more strongly in AUD than outside of it; its weakness is that this includes many
|
| bshanks@30 | 265 areas which don’t have a salient border matching the areal border. The geometric method identifies genes whose salient
|
| bshanks@30 | 266 expression border seems to partially line up with the border of AUD; its weakness is that this includes genes which don’t
|
| bshanks@30 | 267 express over the entire area. Genes which have high rankings using both pointwise and border criteria, such as Aph1a in the
|
| bshanks@30 | 268 example, may be particularly good markers. None of these genes are, individually, a perfect marker for AUD; we deliberately
|
| bshanks@30 | 269 chose a “difficult” area in order to better contrast pointwise with geometric methods.
|
| bshanks@30 | 270 Areas which can be identified by single genes
|
| bshanks@30 | 271 todo
|
| bshanks@30 | 272 Specific to Aim 1 (and Aim 3)
|
| bshanks@30 | 273 Forward stepwise logistic regression todo
|
| bshanks@30 | 274 SVM on all genes at once
|
| bshanks@30 | 275 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
|
| bshanks@30 | 276 surface pixels based on their gene expression profiles. We achieved classification accuracy of about 81%8. As noted above,
|
| bshanks@30 | 277 however, a classifier that looks at all the genes at once isn’t practically useful.
|
| bshanks@30 | 278 The requirement to find combinations of only a small number of genes limits us from straightforwardly applying many
|
| bshanks@30 | 279 of the most simple techniques from the field of supervised machine learning. In the parlance of machine learning, our task
|
| bshanks@30 | 280 combines feature selection with supervised learning.
|
| bshanks@30 | 281 Decision trees
|
| bshanks@30 | 282 todo
|
| bshanks@30 | 283 Specific to Aim 2 (and Aim 3)
|
| bshanks@30 | 284 Raw dimensionality reduction results
|
| bshanks@30 | 285 todo
|
| bshanks@30 | 286 (might want to incld nnMF since mentioned above)
|
| bshanks@30 | 287 _________________________________________
|
| bshanks@30 | 288 they predict area AUD.
|
| bshanks@30 | 289 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
|
| bshanks@30 | 290 shape of area AUD, was calculated, and this was used to rank the genes.
|
| bshanks@30 | 291 85-fold cross-validation.
|
| bshanks@30 | 292 _______________________________________________________________________________________________________
|
| bshanks@30 | 293 PHS 398/2590 (Rev. 09/04) Page 7 ___ Continuation Format Page
|
| bshanks@30 | 294 Principal Investigator/Program Director(Last, First, Middle): Stevens, Charles F.___
|
| bshanks@30 | 295 Dimensionality reduction plus K-means or spectral clustering
|
| bshanks@30 | 296 Many areas are captured by clusters of genes
|
| bshanks@30 | 297 todo
|
| bshanks@30 | 298 todo
|
| bshanks@30 | 299 _______________________________________________________________________________________________________
|
| bshanks@30 | 300 PHS 398/2590 (Rev. 09/04) Page 8 ___ Continuation Format Page
|
| bshanks@30 | 301 Principal Investigator/Program Director(Last, First, Middle): Stevens, Charles F.___
|
| bshanks@30 | 302 Research plan
|
| bshanks@30 | 303 todo amongst other things:
|
| bshanks@30 | 304 Develop algorithms that find genetic markers for anatomical regions
|
| bshanks@30 | 305 1.Develop scoring measures for evaluating how good individual genes are at marking areas: we will compare pointwise,
|
| bshanks@30 | 306 geometric, and information-theoretic measures.
|
| bshanks@30 | 307 2.Develop a procedure to find single marker genes for anatomical regions: for each cortical area, by using or combining
|
| bshanks@30 | 308 the scoring measures developed, we will rank the genes by their ability to delineate each area.
|
| bshanks@30 | 309 3.Extend the procedure to handle difficult areas by using combinatorial coding: for areas that cannot be identified by any
|
| bshanks@30 | 310 single gene, identify them with a handful of genes. We will consider both (a) algorithms that incrementally/greedily
|
| bshanks@30 | 311 combine single gene markers into sets, such as forward stepwise regression and decision trees, and also (b) supervised
|
| bshanks@30 | 312 learning techniques which use soft constraints to minimize the number of features, such as sparse support vector
|
| bshanks@30 | 313 machines.
|
| bshanks@30 | 314 4.Extend the procedure to handle difficult areas by combining or redrawing the boundaries: An area may be difficult to
|
| bshanks@30 | 315 identify because the boundaries are misdrawn, or because it does not “really” exist as a single area, at least on the
|
| bshanks@30 | 316 genetic level. We will develop extensions to our procedure which (a) detect when a difficult area could be fit if its
|
| bshanks@30 | 317 boundary were redrawn slightly, and (b) detect when a difficult area could be combined with adjacent areas to create
|
| bshanks@30 | 318 a larger area which can be fit.
|
| bshanks@30 | 319 Apply these algorithms to the cortex
|
| bshanks@30 | 320 1.Create open source format conversion tools: we will create tools to bulk download the ABA dataset and to convert
|
| bshanks@30 | 321 between SEV, NIFTI and MATLAB formats.
|
| bshanks@30 | 322 2.Flatmap the ABA cortex data: map the ABA data onto a plane and draw the cortical area boundaries onto it.
|
| bshanks@30 | 323 3.Find layer boundaries: cluster similar voxels together in order to automatically find the cortical layer boundaries.
|
| bshanks@30 | 324 4.Run the procedures that we developed on the cortex: we will present, for each area, a short list of markers to identify
|
| bshanks@30 | 325 that area; and we will also present lists of “panels” of genes that can be used to delineate many areas at once.
|
| bshanks@30 | 326 Develop algorithms to suggest a division of a structure into anatomical parts
|
| bshanks@30 | 327 1.Explore dimensionality reduction algorithms applied to pixels: including TODO
|
| bshanks@30 | 328 2.Explore dimensionality reduction algorithms applied to genes: including TODO
|
| bshanks@30 | 329 3.Explore clustering algorithms applied to pixels: including TODO
|
| bshanks@30 | 330 4.Explore clustering algorithms applied to genes: including gene shaving, TODO
|
| bshanks@30 | 331 5.Develop an algorithm to use dimensionality reduction and/or hierarchial clustering to create anatomical maps
|
| bshanks@30 | 332 6.Run this algorithm on the cortex: present a hierarchial, genoarchitectonic map of the cortex
|
| bshanks@30 | 333 _____________________
|
| bshanks@30 | 334 stuff i dunno where to put yet (there is more scattered through grant-oldtext):
|
| bshanks@16 | 335 Principle 4: Work in 2-D whenever possible
|
| bshanks@30 | 336 In anatomy, the manifold of interest is usually either defined by a combination of two relevant anatomical axes (todo), or
|
| bshanks@30 | 337 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
|
| bshanks@30 | 338 the latter case it is curved. If the manifold is curved, there are various methods for mapping the manifold into a plane.
|
| bshanks@30 | 339 The method that we will develop will begin by mapping the data into a 2-D plane. Although the manifold that charac-
|
| bshanks@30 | 340 terized cortical areas is known to be the cortical surface, it remains to be seen which method of mapping the manifold into
|
| bshanks@30 | 341 _______________________________________________________________________________________________________
|
| bshanks@30 | 342 PHS 398/2590 (Rev. 09/04) Page 9 ___ Continuation Format Page
|
| bshanks@30 | 343 Principal Investigator/Program Director(Last, First, Middle): Stevens, Charles F.___
|
| bshanks@30 | 344 a plane is optimal for this application. We will compare mappings which attempt to preserve size (such as the one used by
|
| bshanks@30 | 345 Caret?? ) with mappings which preserve angle (conformal maps).
|
| bshanks@30 | 346 Although there is much 2-D organization in anatomy, there are also structures whose shape is fundamentally 3-dimensional.
|
| bshanks@30 | 347 If possible, we would like the method we develop to include a statistical test that warns the user if the assumption of 2-D
|
| bshanks@30 | 348 structure seems to be wrong.
|
| bshanks@30 | 349 if we need citations for aim 3 significance, http://www.sciencedirect.com/science?_ob=ArticleURL&_udi=B6WSS-4V70FHY-9&_
|
| bshanks@30 | 350 user=4429&_coverDate=12%2F26%2F2008&_rdoc=1&_fmt=full&_orig=na&_cdi=7054&_docanchor=&_acct=C000059602&_version=
|
| bshanks@30 | 351 1&_urlVersion=0&_userid=4429&md5=551eccc743a2bfe6e992eee0c3194203#app2 has examples of genetic targeting to spe-
|
| bshanks@30 | 352 cific anatomical regions
|
| bshanks@30 | 353 —
|
| bshanks@30 | 354 note:
|
| bshanks@30 | 355 do we need to cite: no known markers, impressive results?
|
| bshanks@30 | 356 _______________________________________________________________________________________________________
|
| bshanks@30 | 357 PHS 398/2590 (Rev. 09/04) Page 10 ___ Continuation Format Page
|
| bshanks@26 | 358
|
| bshanks@26 | 359
|
