Introduction Mirko Birbaumer ECCB 2010 26th September 2010 1 / 25 From Vesicle Features to Cellular Phenotypes : Statistical Clustering in Image-based High-Throughput RNAi Screens Mirko Birbaumer birbaumer@imsb.biol.ethz.ch 26th September 2010 Institute of Molecular Systems Biology www.imsb.biol.ethz.ch
Image Analysis Pipeline Mirko Birbaumer ECCB 2010 26th September 2010 2 / 25 Motivation and Problem Description RNA interference experiment : in each well there are thousands of (fixed) cells and the expression of a particular gene is silenced. How a virus enters a cell and how it is transported within a cell. Institute of Molecular Systems Biology www.imsb.biol.ethz.ch
Image Analysis Pipeline Mirko Birbaumer ECCB 2010 26th September 2010 3 / 25 Motivation and Problem Description Observed Vesicle Patterns in RNAi Screen RNAi:Scrambled RNAi: MAPK7 RNAi: CDK8 RNAi: PRKAG1 1 1 1 1 2 2 3 2 2 3 3 3 4 4 4 4 5 A488-Tfn C How can we classify these patterns in an unsupervised manner? Institute of Molecular Systems Biology www.imsb.biol.ethz.ch
Image Analysis Pipeline Mirko Birbaumer ECCB 2010 26th September 2010 4 / 25 Motivation and Problem Description Motivation and Problem Description In the context of RNAi screens, we have observed a plethora of different vesicle patterns Institute of Molecular Systems Biology www.imsb.biol.ethz.ch
Image Analysis Pipeline Mirko Birbaumer ECCB 2010 26th September 2010 4 / 25 Motivation and Problem Description Motivation and Problem Description In the context of RNAi screens, we have observed a plethora of different vesicle patterns How can we quantify these patterns? How can these patterns be distinguished and classified in an unsupervised and automated manner? Institute of Molecular Systems Biology www.imsb.biol.ethz.ch
Image Analysis Pipeline Mirko Birbaumer ECCB 2010 26th September 2010 4 / 25 Motivation and Problem Description Motivation and Problem Description In the context of RNAi screens, we have observed a plethora of different vesicle patterns How can we quantify these patterns? How can these patterns be distinguished and classified in an unsupervised and automated manner? Aim : Find functional modules of genes whose silencing leads to similar vesicle patterns reflecting the function of these genes Institute of Molecular Systems Biology www.imsb.biol.ethz.ch
Image Analysis Pipeline Mirko Birbaumer ECCB 2010 26th September 2010 4 / 25 Motivation and Problem Description Motivation and Problem Description In the context of RNAi screens, we have observed a plethora of different vesicle patterns How can we quantify these patterns? How can these patterns be distinguished and classified in an unsupervised and automated manner? Aim : Find functional modules of genes whose silencing leads to similar vesicle patterns reflecting the function of these genes Furthermore, we want to reveal the regulatory structure between silenced genes in a RNAi screen based on vesicle features Institute of Molecular Systems Biology www.imsb.biol.ethz.ch
Image Analysis Pipeline Mirko Birbaumer ECCB 2010 26th September 2010 5 / 25 Overview of Image Analysis Pipeline Image Analysis Pipeline 1 2 3 4 3,072 Automated Nucleus&cell segmentation SE background wells (994 imaging SVM-filtering of mitotic, subtraction&object controls) at 40x 0.9 NA apoptotic & out-of-focus cells detection; cell annotation 138,240 images (9 sites/well; 305,070 single cells at 20.8x10 6 single vesicles 3 channels; 3 z-planes for Tfn) interphase & in-focus and endosomes SE-detected objects ( ) I max 15’ AF488-Tfn CellTracer DAPI AF488-Tfn Institute of Molecular Systems Biology www.imsb.biol.ethz.ch
Image Analysis Pipeline Mirko Birbaumer ECCB 2010 26th September 2010 6 / 25 Overview of Image Analysis Pipeline Vesicle Features Vesicle Features Feature Description Type v1 Area Size Cargo v2 Integrated intensity content single vesicle or endosome v3 Ellipticity Shape v9 Radius containing 80% Cargo v4 v5 intensity concentration Subcellular v5 Rel. distance to nucleus v8 position v6 Number of neighbours Local v6 v7 within crowding ( ) v2 v1 Number of neighbours Local v7 l within crowding Radius containing 40% Distance to v8 v3 of cell’s vesicles other vesicles w Radius containing 60% Distance to v9 of cell’s vesicles other vesicles v4 Number of vesicles per Endocytic c1 cell area activity single cell Clustering of spatial Overall c2 point pattern (Ripley’s K) pattern Regularity of spatial Overall c3 point pattern (Ripley’s K) pattern Institute of Molecular Systems Biology www.imsb.biol.ethz.ch
Image Analysis Pipeline Mirko Birbaumer ECCB 2010 26th September 2010 7 / 25 Overview of Image Analysis Pipeline Nucleus Classification Nucleus Classification Reality in screens : presence of artefacts or cells are in different mitotic stages GUI for cell classification based on nuclei features and SVM (R package e1071 ); the GUI was written in python ( Tkinter ) in combination with rpy used as an interface to R Institute of Molecular Systems Biology www.imsb.biol.ethz.ch
Image Analysis Pipeline Mirko Birbaumer ECCB 2010 26th September 2010 7 / 25 Overview of Image Analysis Pipeline Nucleus Classification Nucleus Classification Reality in screens : presence of artefacts or cells are in different mitotic stages Aim: we want to keep only in-focus cells at interphase for the analysis CellProfiler provides 52 intensity, texture and shape features of the detected nuclei (feature extraction GUI for cell classification based on nuclei features and with CellProfiler) SVM (R package e1071 ); the GUI was written in python ( Tkinter ) in combination with rpy used as an interface to R Institute of Molecular Systems Biology www.imsb.biol.ethz.ch
Image Analysis Pipeline Mirko Birbaumer ECCB 2010 26th September 2010 7 / 25 Overview of Image Analysis Pipeline Nucleus Classification Nucleus Classification Reality in screens : presence of artefacts or cells are in different mitotic stages Aim: we want to keep only in-focus cells at interphase for the analysis CellProfiler provides 52 intensity, texture and shape features of the detected nuclei (feature extraction GUI for cell classification based on nuclei features and with CellProfiler) SVM (R package e1071 ); the GUI was written in python ( Tkinter ) in combination with rpy used as an interface to Nucleus Classification based on R SVM Institute of Molecular Systems Biology www.imsb.biol.ethz.ch
From Vesicle Features to Cellular Phenotypes Mirko Birbaumer ECCB 2010 26th September 2010 8 / 25 Clustering Approaches Clustering of Vesicle Patterns RNAi:Scrambled RNAi: MAPK7 RNAi: CDK8 RNAi: PRKAG1 1 1 1 1 2 2 3 2 2 3 3 3 4 4 4 4 A488-Tfn 5 C Approach 1 Approach 2 Approach 3 Measurements Vesicle feature Vesicle feature Vesicle feature distributions per cell distributions per cell distributions of all cells PCA (1st 3 PCs) PCA (1st 3 PCs) Vesicle GMM on all cells Mean values Vesicle GMM per cell 7 vesicle subpopulations (see Fig. 3a) Vesicle-averaged Single-cell vectors from single-cell vectors subpopulation fractions Dissimilarity (Euclidean) Kullback-Leibler divergence Dissimilarity (Euclidean) CDK8 2 MDS 2 MDS 0.4 CDK8 MDS ��� CNTRL 2 CDK8 3 30 �� CDK8 MAPK7 2 0.3 3 ��� 2 CNTRL � 20 �� CNTRL 1 4 3 1 CDK8 CDK8 0.2 2 ��� CNTRL 4 MAPK7 5 MAPK7 4 �� 10 0 0.1 ������� 4 � ��� MAPK7 MAPK7 3 2 MAPK7 CDK8 CDK8 4 1 MAPK7 ������� 3 1 4 CDK8 0 MAPK7 3 � CDK8 1 3 ������� ������ 3 CDK8 0.0 4 ��� MAPK7 1 1 5 ������ 4 ������� ������� 4 CDK8 4 ������� 4 MAPK7 -2 ������ 2 �� -10 ��� 1 ������� 1 CNTRL 1 3 ������� ������ -0.1 ���� CNTRL ������ CNTRL 2 1 MAPK7 5 2 3 ������� 1 2 CNTRL 4 MAPK7 -20 ��� CNTRL -0.2 ���� CNTRL 2 CNTRL 2 -4 3 3 ������� �� CNTRL -1,000 -500 0 500 1,000 -10 -5 0 5 -0.4 -0.2 0.0 0.2 0.4 ����� ���� � ��� ���� ��� �� � � ���� ���� ��� ��� ��� Institute of Molecular Systems Biology www.imsb.biol.ethz.ch
From Vesicle Features to Cellular Phenotypes Mirko Birbaumer ECCB 2010 26th September 2010 9 / 25 Clustering Approaches Vesicle Subpopulations Vesicle Subpopulations Biplots and spider graphs reveal properties of vesicle subpopulations. Institute of Molecular Systems Biology www.imsb.biol.ethz.ch
From Vesicle Features to Cellular Phenotypes Mirko Birbaumer ECCB 2010 26th September 2010 10 / 25 Clustering Approaches Vesicle Subpopulations Vesicle Subpopulations Vesicles in Control cells are annotated with colored symbols corresponding to the related subpopulation. Institute of Molecular Systems Biology www.imsb.biol.ethz.ch
Recommend
More recommend
