Hi,

My group are planning to conduct a 15k compound high content biology screen and I'm looking into options for analysing the image data.  We'd probably just be doing a simple cell count and maybe cell size to start with.  Although we have commerical software to do this (Definiens), I'm not sure how well our set-up will scale given the hardware it's on.  Another option is to use our institute's NGS compute cluster, and analyse the data using the Bioconductor package EBImage which I think would scale much better.

In terms of learning resources there's the package vignette and the following tutorial from CSAMA2015, but are there any additional workflow papers or other resources that might be helpful to me in setting this up?

http://www.bioconductor.org/help/course-materials/2015/BioC2015/BioC2015Oles.html

Thanks,

Phil

Hi Phil

It's a good idea to do this with EBImage. We've used it for several imaging-based HT screens. The vignette of the paper  "A Map of Directional Genetic Interactions in a Metazoan Cell" by Bernd Fischer et al. generates all figures in the paper. It has a section on the HT image processing per se (as well as the downstream QA/QC, dimension reduction etc.): http://bioconductor.org/packages/release/data/experiment/html/DmelSGI.html

There is also a draft version of a teaching chapter on image data analysis that heavily relies on EBImage, and which I am happy to share (caveat - it's really a draft, forgive the bumps): http://www.huber.embl.de/users/whuber/.useR-book/Draft-Image-Analysis.pdf

Let us know about your experiences or questions that come up. Pull requests or suggestions are also welcome. It would also be interesting to hear what others on this forum use? E.g. there is also http://www.cellcognition.org/ (focus on time lapse imaging; interoperable with R through the cellh5 data exchange format (HDF5-based, http://www.cellh5.org ), and http://ilastik.org/ uses a clever interactive learning approach that maybe helpful if the segmentation is difficult.

Wolfgang

Dear Phil,

you can also have a look at the supplemental data packages accompanying the following papers. Each of these packages' vignette contains a description of the image analysis pipeline.

• Mapping genetic interactions in human cancer cells with RNAi and multiparametric phenotyping by Laufer et al. (2013)
• Directional tissue migration through a self-generated chemokine gradient by Donà et al. (2013)
• A chemical-genetic interaction map of small molecules using high-throughput imaging in cancer cells by Breinig et al. (2015)
• An optogenetic method to modulate cell contractility during tissue morphogenesis by Guglielmi et al. (2015)

Cheers,
Andrzej

Some other useful links for posterity:

Earl Glynn's RNotes blog: http://earlglynn.github.io/RNotes/package/EBImage/index.html

Slideset from Wolfgang Huber: http://mlpm.eu/static/media/uploads/mlpm13_slides_huber01.pdf

Machine Learning in cell biology - teaching computers to read phenotypes: http://jcs.biologists.org/content/joces/126/24/5529.full.pdf