Martin Stöter HT - Technology Development Studio (TDS), KNIME workshop the HC-Screening Unit at the MPI-CBG March 2nd – 4th 2011, Zürich [email protected] Screen Mining with KNIME A user-friendly framework for high throughput/content data analysis Outline - Our challenges with High-Content Screening (HCS) data - HCS Tools (Community nodes) - Example Workflows (pseudo demo) - with HCS Tools - R templates - Dose Response with variables Martin Stöter, MPI-CBG, Dresden, Germany 2 Technology Development Studio (TDS) MPI-CBG, Dresden, Germany Screening facility for academic laboratories Provide full service for automation and cell-based screens, RNAi and chemical screens Equipment: liquid handling robots, drop dispensers, plate washers, plate readers, High Content Screening platforms Martin Stöter, MPI-CBG, Dresden, Germany 3 Is Data Analysis a Bottleneck in HCS? Complex Experiments Lots of data (too much for Excel) Fancy data analysis / mining Scientists Many scientists, but few data analysts Sometimes different languages Data analysis is often a bottleneck! Data analyst Konstanz Information Miner (KNIME) Open source data analysis platform Modular data pipelining concept Courtesy of Holger Brandl, MPI-CBG 4 Template Scripts for KNIME Excel, tables, R, Python, graphical user Matlab, Java, interfaces programming Scientist Data analyst Solution Data analysts write template scripts AND users access these using a graphical interface! R Other scripting languages Local or R server, easy to update & to edit Groovy, Java snippets, JPython, Perl Visualizations (boxplots, histograms, profiles, …) Matlab Other KNIME functionalities Open in Matlab, Matlab snippet, Mattlab plot Python Chemoinformatics Open in Python, Python snippet, Python plot Bioinformatics Data mining Courtesy of Holger Brandl, MPI-CBG 5 High-Content Screening (HCS) data Data generation - Cells (RNAi, compounds) - Microscopy -> images - Image analysis - Cell features/parameters -> well data Tasks/problems - Read data from various sources SQL database, XML, Excel, various .csv … - Screening specific statistics - Screening specific utilities - Data mining, visualization 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 A B DMSO DMSO DMSO C 0.001 DMSO DMSO 0.001 D 10 DMSO DMSO 10 E 10 DMSO DMSO 10 F 3 DMSO DMSO 3 G 3 DMSO DMSO 3 H 1 DMSO DMSO 1 I 1 DMSO DMSO 1 J 0.3 DMSO DMSO 0.3 K 0.3 DMSO DMSO 0.3 L 0.1 DMSO DMSO 0.1 M no AB no AB 0.1 DMSO DMSO 0.1 N no AB no AB 0.1 DMSO DMSO 0.1 O DMSO DMSO P Martin Stöter, MPI-CBG, Dresden, Germany 6 HCS HCS Tools for KNIME Tools Data Import Image Analysis (Opera, Operetta, Cell Profiler, MotionTracking) Plate Readers (Envision, GeniusPro, MSD SectorImager) Normalization Percent-of-control (POC), Normalized percent inhibition (NPI) Z-score, B-score Optional: robust statistics (Median + MAD) Select wells to normalize (controls, samples) Quality Control CV, Z‘, Multivariate Z‘, SSMD Screen Mining Annotation of screen data from database Dose Response (IC50) Utilities Handle barcodes & wells, join layouts Visualization Plate Viewer -> heatmaps, brows wells, ... Mondiran, R templates Martin Stöter, MPI-CBG, Dresden, Germany 7 HCS Tools: Standardized Data Format - Different readers nodes to shape a common data structure - Enforce standardization of data format - Lower the knowledge entry barrier for new users HCS Tools: Barcode Standard Regular expression for interpretation of barcode: Standardized table structure -> connection to our TDS compound database (?<libplatenumber>[0-9]{3})(?<projectcode>[A-z]{2})(?<date>[0-9]{6})(?<replicate>[A-z]{1}) Configurable in Preferences -> KNIME -> HCA Tools HCS Tools: Annotate Experiment Excel is (still) the tool of choice for assay development Join Layout node is Excel Reader for defined spread sheet Plate format with multiple well attributes (1 plate layout -> 1 column in KNIME) KNIME Workflow Example 1 Plate Viewer: Heatmaps of entire Screen 179 plates x 384wells = ~70.000 data points times x parameters Martin Stöter, MPI-CBG, Dresden, Germany 12 Workflow Example 2 Martin Stöter, MPI-CBG, Dresden, Germany 13 Workflow Example 2: Clustering Martin Stöter, MPI-CBG, Dresden, Germany 14 Workflow Example 2 Martin Stöter, MPI-CBG, Dresden, Germany 15 Workflow Example 2: Dose Response node Martin Stöter, MPI-CBG, Dresden, Germany 16 Using flow variables in R scripts Dose Response node looks like KNIME, but actually it is R!!! Martin Stöter, MPI-CBG, Dresden, Germany 17 Summary • HCS Tools provides a very useful functionality for HCS specific applications in KNIME • The scripting template environment allows to use a GUI for configuration of scripts. • R scripts from templates can be modified and customized in any node and support flow variables. How to get the HCS Tools & Scripting nodes? MPI-CBG http://www.mpi-cbg.de/facilities/profiles/software-engineering/hcs-tools.html HCS Tools KNIME community website http://tech.knime.org/hcs-tools Link KNIME community update site http://tech.knime.org/update/community-contributions/nightly KNIME -> Help -> Install Software -> Add… Paste update link and select node packages to be installed …and the R templates? -> links comes automatically with Scripting nodes, but can be configured in KNIME preferences. Martin Stöter, MPI-CBG, Dresden, Germany 19 Acknowledgements Software Development / TDS team (MPI-CBG) Bioinformatics Facility (MPI-CBG) Marc Bickle Felix Meyerhofer Holger Brandl Cordula Andre Claudia Möbius Rico Barsacchi Antje Niederlein HCS Sara Christ Nadine Tomschke Tools Milan Esner Jan Wagner Annett Lohmann KNIME Michael Berthold and the KNIME team 20 Thank you for your attention! Martin Stöter, MPI-CBG, Dresden, Germany 21 .