Computational Biology, Part 23 Image-based Cell ModelsGenerative models of subcellular patternsLAMP2 patternNuclear Shape - Medial Axis ModelSynthetic Nuclear ShapesWith added nuclear textureCell Shape Description: Distance RatioGenerationModeling Vesicular OrganellesObject PositionsSynthesized ImagesEvaluation of synthesized imagesModel DistributionGeneration ProcessGenerating Multiple Distributions for SimulationsCombining Models for Cell SimulationsFinal wordModels that consider compartment geometrySlide 19Slide 20Slide 21Making a compartment map for Virtual Cell from a fluorescence microscope imageSlide 23Make contiguous cytoplasm image by averaging weak autofluorescenceCombine with image of pixels with positive lysosomal stainingResulting imageComputational Biology, Part 23Image-based Cell ModelsComputational Biology, Part 23Image-based Cell ModelsRobert F. MurphyRobert F. MurphyCopyright Copyright  2005-2008. 2005-2008.All rights reserved.All rights reserved.Generative models of subcellular patterns Generative models of subcellular patternsLAMP2 patternLAMP2 patternNucleusCell membraneProteinNuclear Shape - Medial Axis ModelNuclear Shape - Medial Axis ModelRotateMedial axisRepresented by two curvesthe medial axiswidth along the medial axiswidthSynthetic Nuclear ShapesSynthetic Nuclear ShapesWith added nuclear textureWith added nuclear textureCell ShapeDescription: Distance RatioCell ShapeDescription: Distance Ratiod1d2221dddr+=Capture variation as a Capture variation as a principal components principal components modelmodelGenerationGenerationModeling Vesicular OrganellesModeling Vesicular OrganellesOriginal Filtered Fitted GaussiansObject PositionsObject Positionsd1d2212dddr+=Synthesized ImagesSynthesized ImagesLysosomes EndosomesEvaluation of synthesized imagesEvaluation of synthesized imagesClassification of synthesized images by a classifier trained on real images. Classification based on features that made 94% of real images distinguishableModel DistributionModel DistributionGenerative models provide better way of Generative models provide better way of distributing what is known about “subcellular distributing what is known about “subcellular location families” (or other imaging results, location families” (or other imaging results, such as illustrating change due to drug such as illustrating change due to drug addition)addition)Have initial XML design for capturing the Have initial XML design for capturing the models for distributionmodels for distributionHave portable tool for generatingHave portable tool for generatingimages from the modelimages from the modelGeneration ProcessGeneration ProcessProteinCell ShapeNuclear ModelXMLGenerating Multiple Distributions for SimulationsGenerating Multiple Distributions for SimulationsProteinCell ShapeNuclear ModelXMLSimulation 1Simulation 2Simulation 3ConclusionsCombining Models for Cell SimulationsCombining Models for Cell SimulationsProtein 1Cell ShapeNuclear ModelProtein 2Cell ShapeNuclear ModelProtein 3Cell ShapeNuclear ModelXMLSimulationShared Nuclear and Cell ShapeFinal wordFinal wordGoal of automated image interpretation Goal of automated image interpretation should not beshould not beQuantitating intensity or colocalizationQuantitating intensity or colocalizationMaking it easier for biologists to see what’s Making it easier for biologists to see what’s happeninghappeningGoal should be generalizable, verifiable, Goal should be generalizable, verifiable, mechanistic models of cell organization and mechanistic models of cell organization and behavior behavior automatically automatically derived from imagesderived from imagesModels that consider compartment geometryModels that consider compartment geometryVirtual Cell facilitated Ca-diffusion model Virtual Cell facilitated Ca-diffusion model from tutorialfrom tutorialhttp://www.nrcam.uchc.edu/login/facil_ca_dhttp://www.nrcam.uchc.edu/login/facil_ca_dif.pdfif.pdfMaking a compartment map for Virtual Cell from a fluorescence microscope imageMaking a compartment map for Virtual Cell from a fluorescence microscope image Start from a Start from a fluorescence fluorescence microscope microscope image of a image of a lysosomal lysosomal protein protein (LAMP-2)(LAMP-2)Making a compartment map for Virtual Cell from a fluorescence microscope imageMaking a compartment map for Virtual Cell from a fluorescence microscope imageUse Matlab to create an image with values Use Matlab to create an image with values of zero for background, one for cytoplasm, of zero for background, one for cytoplasm, and two for lysosomesand two for lysosomesAssume that the autofluorescence in the Assume that the autofluorescence in the lysosome image is sufficient to find a region lysosome image is sufficient to find a region corresponding to the cytoplasmcorresponding to the cytoplasmMake contiguous cytoplasm image by averaging weak autofluorescenceMake contiguous cytoplasm image by averaging weak autofluorescenceimg=imread('r06aug97.h4b4.13--1---2.dat.png');img=imread('r06aug97.h4b4.13--1---2.dat.png');a=double(img);a=double(img);b=(a-min(min(a)))./(max(max(a))-min(min(a)));b=(a-min(min(a)))./(max(max(a))-min(min(a)));H=fspecial('average',13);H=fspecial('average',13);c=imfilter(b,H,'replicate');c=imfilter(b,H,'replicate');d=im2bw(c,0.004);d=im2bw(c,0.004);imshow(d);imshow(d);max(max(d))max(max(d))Combine with image of pixels with positive lysosomal staining Combine with image of pixels with positive lysosomal staining e=im2bw(b,graythresh(b));e=im2bw(b,graythresh(b));imshow(e);imshow(e);f=d + e;f=d + e;imshow(f,[0 2]);imshow(f,[0 2]);g=uint8(f);g=uint8(f);imwrite(g,'geomap.tif','TIF','Compression','none');imwrite(g,'geomap.tif','TIF','Compression','none');Resulting image Resulting