Automated Data Pipelines & Intelligent Microscopy

Size: px

Start display at page:

Download "Automated Data Pipelines & Intelligent Microscopy"

Ann Allen
5 years ago
Views:

1 Galaxy Workshop Freiburg University Automated Data Pipelines & Intelligent Microscopy RNAi Screening Facility BioQuant, Heidelberg University Jürgen Reymann

2 Scientific Infrastructure Automated Data Pipelines Intelligent Automated Imaging

3 BioQuant Center for Systems Biology Quantitative Analysis of Molecular and Cellular Biosystems Interaction of viral and cellular Systems Modeling Platform Technology Platform Large Scale Data Facility

500 TB p.a. Large Scale Data Facility M.

4 Data Analytics Infrastructure Super resolution microscopy (>100 TB p.a.) Light sheet microscopy (>100 TB p.a.) hard drives 6 PB total capacity 500 TB p.a. Large Scale Data Facility M. Hemberger 5 PB p.a. 1 PB p.a. J. Eils, C. Lawerenz

5 RNAi Screening Facility Support

RNAi Solid Phase Reverse Transfection sirna mirna cdna complexed with tranfection solution Gene specific degradation of target mrna in the cells

6 RNAi Solid Phase Reverse Transfection sirna mirna cdna complexed with tranfection solution Gene specific degradation of target mrna in the cells Loss of function phenotype INCENP -phenotype sirna spot distribution on cell array Spot diameter: 400µm Ziauddin J. and Sabatini D.M., Nature 411, (2001)

genome) 384er head printing robot Erfle et al.

7 Assay Automation 8er head printing robot 384er well plates LabTeks (384er cell arrays) Cell arrays (whole genome) 384er head printing robot Erfle et al., JBS, 2008 Erfle et al., Nat. Prot. (2), 2007 Reymann et al., BioTech, 2009

8 Object Scanning 2D low resolution: Fully automated [ 10x/20x objective lens ] 1 scan field: 1-3s (1 channel auto-focusing) 1000 scan fields: 1 hour (3 channels auto-focusing stage moving) Wide field [ Olympus IX81 ScanR ] 3D high resolution: Fully automated [ 60x/63x objective lens ] 1 scan field: 30s (1 channel 30 optical sections auto-focusing) 1000 scan fields: 1 day (3 channels auto-focusing stage moving) Confocal [ Leica TCS SP5 ] 2D super resolution: Semi-automated [ 63x/100x objective lens ] 1 scan field: 3-10min (1 channel) 1000 scan fields: - Confocal [ Leica TCS SP5 ] Integrated: Wide field Super resolution

9 Screening Workflow & Timescales

10 Scientific Infrastructure Automated Data Pipelines Intelligent Automated Imaging

11 Biological Question Data Mining Assay Development Imaging Experiment Setup

12 Computing power Software platforms Algorithms Basic user availability Data Mining Biological Question Preparation methods Labeling strategies Fluorophores Parallel assays Assay Development Automation Correlative Throughput Resolution Feedback options Imaging Experiment Setup

13 Computing power Software platforms Algorithms Basic user availability Biological Question Preparation methods Labeling strategies Fluorophores Parallel assays Data Mining Increase of complexity Increase of information content Extraction / Interpretation of information Assay Development Automation Correlative Throughput Resolution Feedback options Imaging Experiment Setup

14 Computing power Software platforms Algorithms Basic user availability Biological Question Preparation methods Labeling strategies Fluorophores Parallel assays Data Mining Increase of complexity Increase of information content Extraction / Interpretation of information Assay Development Automation Correlative Throughput Resolution Feedback options Imaging Experiment Setup

15 Data Pipelines Screening Workflow :: Biological Assay Genome wide RNAi screening platform to identify host factors involved in HIV-1 (HCV, DV, AAV, PV) replication Boerner K. et al Biotechnol. J, 2010

16 Data Pipelines Information Workflow :: Data Mining Visualisation of raw data, meta data, intermediate and ultimate results Transfection protocol sirna library Sample preparation Assay automation Plate layouts Usability for basic user Data acquisition High-throughput High-content Correlative Data storage Image Processing Quality control Normalisation Segmentation Features Modular structure Statistics Normalisation (per plate/assay) Spatial normalisation Hit calling High-throughput capable, i.e. handling of large data sets, parallel computing, serverbased functionalities Bioinformatics Data integration Processes Pathways Interaction networks Extensive integrations Libraries, format handling, data manipulation, Screen meta data Screen meta data Image raw data Image meta data Analysis results Analysis results Analysis results Usability / Easy Integration of established algorithms Possibility to link to external data bases [ Visualisation (plate viewer, image viewer, plots) Tables Meta data ] Base Data

17 Data Pipelines Information Workflow :: Data Mining [ ] Visualisation of raw data, meta data, intermediate and ultimate results Transfection protocol sirna library Sample preparation Assay automation Plate layouts Usability for basic user Data acquisition High-throughput High-content Correlative Data storage Image Processing Quality control Normalisation Segmentation Features Modular structure Statistics Normalisation (per plate/assay) Spatial normalisation Hit calling High-throughput capable, i.e. handling of large data sets, parallel computing, serverbased functionalities Bioinformatics Data integration Processes Pathways Interaction networks Extensive integrations Libraries, format handling, data manipulation, Screen meta data Screen meta data Image raw data Image meta data Analysis results Analysis results Analysis results Usability / Easy Integration of established algorithms Possibility to link to external data bases [ Visualisation (plate viewer, image viewer, plots) Tables Meta data ] Base Data

18 Data Pipelines Information Workflow :: Data Mining [ ] Modular structure Visualisation of raw data, meta data, intermediate and ultimate results Usability for basic user High-throughput capable, i.e. handling of large data sets, parallel computing, serverbased functionalities Open-source / Platform independant Extensive integrations Libraries, format handling, data manipulation, Sustainability Supported by a large (developer) community / Usage of universal standards Usability / Easy Integration of established algorithms Possibility to link to external data bases

19 Data Pipelines Information Workflow :: Data Mining Integrations / Libraries Data Handling / Data Views ImageJ / ImageJ2 ImgLib2 (Fiji) WEKA R SVM s Clustering Classifiers Network Mining Normalisations Microscope Control Target identification FeedBack Data Base File reader / writer BioFormats (image data) OME-XML (image data) XML support Histograms / Plots Image / Table / Plate viewer Data manipulation Loop control OMERO

20 Data Pipelines Information Workflow :: Data Mining Workflow Repository Workflow Configuration Node Description Node Repository

21 Data Pipelines Information Workflow :: Data Mining Workflow Configuration Nuclei registration

22 Data Pipelines Information Workflow :: Data Mining Workflow Configuration Nuclei registration

23 Data Pipelines Information Workflow :: Data Mining Workflow Configuration Nuclei registration

24 Data Pipelines Information Workflow :: Data Mining Workflow Configuration Nuclei registration

25 Data Pipelines Information Workflow :: Data Mining Workflow Configuration Nuclei registration 02/10/2014 Galaxy Workshop Freiburg University

26 Data Pipelines

27 Data Pipelines Experiment Meta Data 96er head pipetting robot Well plates Printing robots 384 cell arrays (LabTeks) Whole genome cell arrays

28 Data Pipelines Image Raw Data Wide field [ 3x Olympus IX81 ScanR ] Confocal [ 2x (exp.) Leica TCS SP5 ] Integrated: Wide field Super resolution Spinning disc [ Perkin Elmer Opera LX ]

29 Data Pipelines Image Processing 2D high-throughput images Nuclei - NO phenotypic penetration Wide field Object-registration

30 Data Pipelines Statistics 2D high-throughput images Nuclei - NO phenotypic penetration sirna INCENP phenotypes Wide field Object-registration Hit-calling

31 Data Pipelines Statistics n = 8 Rel. distance of object to INCENP feature space Positive classified (Rel. to Incenp feature space) Negative classified (Rel. to Incenp feature space) Object

32 Data Pipelines Statistics n = 8 Rel. distance of object to INCENP feature space Positive classified (Rel. to Incenp feature space) n = 32 Erb n = 35 KIF11 n = 8 INCENP n = 35 PLK Negative classified (Rel. to Incenp feature space) Object

33 Data Pipelines Workflow Repository

34 Scientific Infrastructure Automated Data Pipelines Intelligent Automated Imaging

35 Computing power Software platforms Algorithms Basic user availability Data Mining Biological Question Preparation methods Labeling strategies Fluorophores Parallel assays Assay Development Automation Correlative Throughput Resolution Feedback options Imaging Experiment Setup

36 Computing power Software platforms Algorithms Basic user availability Data Mining Biological Question Preparation methods Labeling strategies Fluorophores Parallel assays Assay Development integrated Data Analysis Automation Correlative Throughput Resolution Feedback options i Imaging Experiment Setup

37 Computing power Software platforms Algorithms Basic user availability Data Mining Biological Question Preparation methods Labeling strategies Fluorophores Parallel assays Assay Development integrated Data Analysis Automation Correlative Throughput Resolution Feedback options i Imaging Experiment Setup

General Considerations Why Automated DAQ? Screening Large scale experiment number (up to genome wide) e.g.: 20.273 human genes *3 sirnas = 60.

38 General Considerations Why Automated DAQ? Screening Large scale experiment number (up to genome wide) e.g.: human genes *3 sirnas = experiments Automated DAQ in general Automation NOT necessarily restricted to screening procedures Biological variability Increase amount of data / # Objects to be scanned Statistical relevance Reproducibility Increase image quality (e.g. quality controls) Increase information content (e.g. correlative microscopy / time lapse) Simplify data acquisition procedures

to screening procedures Biological variability Increase amount of data / # Objects to be

39 General Considerations Why Automated DAQ? Screening Large scale experiment number (up to genome wide) e.g.: human genes *3 sirnas = experiments Junk data Automated DAQ in general Automation NOT necessarily restricted to screening procedures Biological variability Increase amount of data / # Objects to be scanned Statistical relevance Reproducibility Increase image quality (e.g. quality controls) Increase information content (e.g. correlative microscopy / time lapse) Simplify data acquisition procedures Junk data Hits

40 General Considerations FeedBack -driven DAQ n-dim feature space of observables [ cells ]

41 General Considerations FeedBack -driven DAQ n-dim feature space of observables [ cells ] Shape Intensity Osterwald S. et al, Biotechnology J Co-localisations

42 General Considerations FeedBack -driven DAQ n-dim feature space of observables [ cells ] Shape PROBLEM: General measurement provides cloud of data points containing every feature! Intensity BUT: Every measurement aims at collecting the information content of specified features. Osterwald S. et al, Biotechnology J Co-localisations

43 General Considerations FeedBack -driven DAQ n-dim feature space of observables [ cells ] Shape PROBLEM: General measurement provides cloud of data points containing every feature! Intensity BUT: Every measurement aims at collecting the information content of specified features. Co-localisations Pre-Processing Identify objects of interest FeedBack to measuring system Acquire only objects of interest

General Considerations FeedBack -driven DAQ n-dim feature space of observables [ cells ] Shape Pure screening time: 6 months PROBLEM: Junk data: General measurement ~40% provides cloud of data points

44 General Considerations FeedBack -driven DAQ n-dim feature space of observables [ cells ] Shape Pure screening time: 6 months PROBLEM: Junk data: General measurement ~40% provides cloud of data points containing every feature! Intensity BUT: Every measurement aims at collecting the information content of specified features. Osterwald Osterwald S. et al, S. et Biotechnology al, J. J Co-localisations objects of interest Pre-Processing Identify objects of interest FeedBack to measuring system Acquire only objects of interest

45 Automated Imaging Combine/Link DAQ IP Transfection protocol Sample preparation Data acquisition Image Processing Statistics Bioinformatics sirna library Assay automation Plate layouts High-throughput High-content Correlative Data storage Quality control Normalisation Segmentation Features Normalisation (per plate/assay) Spatial normalisation Hit calling Data integration Processes Pathways Interaction networks Use information from Image Pre- Processing in order to optimise / upgrade DAQ

46 Automated Imaging Combine/Link DAQ IP Transfection protocol Sample preparation Data acquisition Image Processing Statistics Bioinformatics sirna library Assay automation Plate layouts High-throughput High-content Correlative Data storage Quality control Normalisation Segmentation Features Normalisation (per plate/assay) Spatial normalisation Hit calling Data integration Processes Pathways Interaction networks Use information from Image Pre- Processing in order to optimise / upgrade DAQ Junk data Costs Feasibility of experiments Bleaching Trigger (time lapse experiments) Pick only objects of interest Quality / Information content of image raw data Screening time Pick representative cells for high/super -resolution

47 Automated Imaging Data acquisition High-throughput High-content Correlative Data storage Image Pre-Processing Targets Trigger Hit calling Image Processing Quality control Normalisation Segmentation Features

Pre-Processing Targets Trigger Hit calling Objects of

48 Automated Imaging Data acquisition High-throughput High-content Correlative Data storage Image Pre-Processing Targets Trigger Hit calling Objects of interest Image Processing Quality control Normalisation Segmentation Features

49 Scientific Infrastructure Data Pipelines & Data Analysis Cross-Platform Solution

50 Cross-Platform Solution Idea: Acquire images of the same sample/target at multiple systems, in order to combine the advantages of different microscopic techniques. Examples:

51 Cross-Platform Solution Problem: Transfer System Coordinate transfer by 1.) Reference markers 2.) (SIFT) image registration

52 Cross-Platform Solution Problem: Transfer System Coordinate transfer by 1.) Reference markers 2.) (SIFT) image registration High-resolution Leica SP5 10 µm Image Processing Target identification Super-resolution dstorm 1 µm dstorm setup: Mike Heilemann Gunkel et al., Histochem & Cell Biol 2014

High-resolution screening Super-resolution screening C Collab.

53 Cross-Platform Solution Integrated Correlative Microscopy High-throughput screening [ Target identification ] High-resolution screening Super-resolution screening C Collab.: Mike Heilemann, Fred Hamprecht, Vytaute Starkuviene Flottmann et al., Biotechniques 2013

54 Scientific Infrastructure Data Pipelines & Data Analysis Automated Microscopy

55 Automated Microscopy Leica Microsystems Computer Aided Microscopy (CAM) -Interface Enables to communicate with the microscope control software [ Retrieve / Send commands ] CAM Data storage (Pre-Screen) Image raw data Targets / Trigger Collab.:

56 Automated Microscopy Integrated Data Pipelines Collab.:

57 Automated Microscopy Integrated Data Pipelines Pre-Screen Configuration 1 2D low resolution imaging of full sample CAM Collab.:

58 Automated Microscopy Integrated Data Pipelines Pre-Screen Configuration 1 2D low resolution imaging of full sample CAM Image Pre-Processing Identify objects of interest (candidates) Collab.:

59 Automated Microscopy Integrated Data Pipelines Pre-Screen Configuration 1 2D low resolution imaging of full sample CAM Image Pre-Processing Identify objects of interest (candidates) Collab.:

60 Automated Microscopy Integrated Data Pipelines Target Screen Configuration 2 3D high resolution imaging of candidates CAM High resolution 3D information of sub-nuclear objects Collab.:

73 Automated Microscopy Highly accurate positioning of targets in the center of field of views Experiment setup Overall sample size: 8 * 12 = 96 scan fields Standard DAQ Config 2» Screening time: ~70h» junk data / less targets (NOT centered!) Automated Imaging Pre-Screen job1 Target Screen job2» Screening time: ~7h» 146 target regions (in center of field of view!) Overall acquisition time: 7 h 17 min Decreased overall screening time (factor 10) Increased number and quality (centered) of acquired target regions

74 Automated Microscopy Highly accurate positioning of targets in the center of field of views Experiment setup Overall sample size: 8 * 12 = 96 scan fields Standard DAQ Config 2» Screening time: ~70h» junk data / less targets (NOT centered!) Automated Imaging Pre-Screen Config 1 Target Screen Config 2» Screening time: ~7h» 146 target regions (in center of field of view!) Overall acquisition time: 7 h 17 min Decreased overall screening time (factor 10) Increased number and quality (centered) of acquired target regions

Automated Microscopy Some Numbers Costs confocal microscopy: 40 per hour Decreased overall screening time (factor 10) Increased number and quality (centered) of acquired target regions Overall sample

75 Automated Microscopy Some Numbers Costs confocal microscopy: 40 per hour Decreased overall screening time (factor 10) Increased number and quality (centered) of acquired target regions Overall sample size: 96 positions Standard DAQ job2» Screening time: ~70h Automated Imaging Pre-Screen job1 Target Screen job2» Screening time: ~7h Overall sample size: positions Standard DAQ job2» Screening time: 6 months» ~40% junk data Automated Imaging Pre-Screen job1 Target Screen job2» Screening time: 3,6 months ~ junk data

76 Perspectives Usability for basic user Data Analysis Data Management Visualisation Image Processing Statistics Bioinformatics Data storage File access File formats (Image) Raw data (Intermediate) Results Hists / Plots / Viewer Design automated workflow solutions Integrated Data Analysis Meta Data Data Bases Workflow repository Quality control Hit/Event identification Optical configuration Experiment Data Analysis Annotations Internal access External access Correlations Biological Question Data Acquisition Data Analysis Modeling Assay automation Plate layouts High-throughput High-content Correlative Data storage Quality control Normalisation Segmentation Features Normalisation (per plate/assay) Spatial normalisation Hit calling

77 Perspectives Usability for basic user Data Analysis Data Management Visualisation Image Processing Statistics Bioinformatics Data storage File access File formats (Image) Raw data (Intermediate) Results Hists / Plots / Viewer Design automated workflow solutions Integrated Data Analysis Meta Data Data Bases Workflow repository Quality control Hit/Event identification Optical configuration Experiment Data Analysis Annotations Internal access External access Correlations Biological Question Assay automation Plate layouts Data Acquisition i High-throughput High-content Correlative Imaging Data storage Data Analysis Quality control Normalisation Segmentation Features Modeling Normalisation (per plate/assay) Spatial normalisation Hit calling Integrated Data Analysis

78 Acknowledgements Holger Erfle Nina Beil Jürgen Beneke Ruben Bulkescher Manuel Gunkel Tautvydas Lisauskas Bastian Schumacher Michael Berthold Christian Dietz Thomas Gabriel Martin Horn Michael Zinsmeier Frank Sieckmann Constantin Kappel Werner Knebel Mike Heilemann Benjamin Flottmann

79 Perspectives Bioinformatics How could Galaxy help?

80 Perspectives Bioinformatics How could Galaxy help? How could Galaxy contribute?

Multiple Usage of KNIME in a Screening Laboratory Environment

Multiple Usage of KNIME in a Screening Laboratory Environment KNIME UGM Zürich, 02.02.2012 Marc Bickle HT-TDS, MPI-CBG Outline Presentation of TDS Our problem: large complex datasets KNIME as data mining