CellProfiler

Top

00:00:15.00 Hi, I'm Anne Carpenter from the Broad Institute,
00:00:17.08 and one of the leaders of the CellProfiler project.
00:00:19.18 Today, I'd like to introduce you to CellProfiler.
00:00:22.00 It's free and open-source software for image analysis
00:00:24.18 that can identify and measure biological entities and images;
00:00:27.06 process large images at any scale,
00:00:29.11 from a small-scale experiment to a large one;
00:00:31.21 and export data for further analysis.
00:00:34.12 First, some basics about CellProfiler.
00:00:36.09 It was designed by a biologist -- that's me.
00:00:38.09 I wrote the first version many years ago
00:00:40.13 to fill the gap between the latest computational methods
00:00:42.19 and everyday bench biologists.
00:00:44.08 It's free.
00:00:45.17 And because it's open-source, you can write your own modules if you need to.
00:00:48.11 It runs on Windows, Mac, and Linux,
00:00:50.22 interfaces with many other popular bioimage analysis software,
00:00:53.22 and it reads more than 180 microscopy file formats
00:00:57.00 thanks to Bio-Formats.
00:00:58.18 And by many metrics, CellProfiler is very popular
00:01:00.26 and well-loved among biologists.
00:01:03.08 The best place to start with CellProfiler
00:01:05.04 is to find an example pipeline,
00:01:06.24 either from a paper that you've read where CellProfiler was used;
00:01:09.08 from the online question-and-answer forum, forum.sc;
00:01:13.13 or from CellProfiler's example page, and some of these examples are shown here.
00:01:18.21 Once you've loaded a pipeline into CellProfiler,
00:01:20.14 you can start adjusting it to customize to your biological problem.
00:01:24.10 The main goal, of course, is to identify and measure
00:01:27.04 some kind of biological entities,
00:01:28.16 whether it's cells or colonies or synapses and so on.
00:01:31.09 The decisions to be made are,
00:01:33.03 which structures or regions or compartments do you want to identify?,
00:01:36.04 how to identify them?,
00:01:37.13 and finally, which features to measure?
00:01:39.01 And as you put the modules together in CellProfiler,
00:01:41.03 these are the kinds of questions you'll be asking yourself.
00:01:44.07 You mix and max... match these modules to build a pipeline,
00:01:47.12 or a workflow, that accomplishes your task.
00:01:49.25 Now, some of these steps are very trivial,
00:01:51.13 for example splitting the colors of a multi-channel image.
00:01:53.25 Some of the steps you might not have thought of before,
00:01:56.03 but you... it's very important to do things like
00:01:58.19 correcting the illumination in order to improve the quality of the quantification
00:02:01.12 that you get from your images.
00:02:03.22 Once you've preprocessed your images to your liking,
00:02:06.04 you can put in modules that identify structures and compartments of interest,
00:02:08.23 such as nuclei or the cell borders,
00:02:10.16 as shown here.
00:02:12.04 This is often the challenging part of setting up an image analysis workflow,
00:02:15.09 but there's lots of tools and tips to give you...
00:02:17.27 give you a helping hand there.
00:02:20.09 Some of the kinds of parameters that you'll need to adjust in that process
00:02:23.28 include adjusting settings for determining the foreground
00:02:26.26 versus the background.
00:02:28.08 So, here's an example where it's a little too strin...
00:02:30.08 a little too lenient, and then a little too stringent,
00:02:31.29 and finally just right.
00:02:33.23 There's other settings that allow you to
00:02:36.14 decide the appropriate splitting versus merging for some objects.
00:02:39.06 So, you can see there's four nuclei here
00:02:40.27 that are shown all stuck together.
00:02:42.14 We adjusted the settings until those were separated properly,
00:02:45.13 using the various settings within the modules.
00:02:48.24 And once you've identified your structures of interest,
00:02:50.28 it's actually very straightforward to measure their properties.
00:02:53.28 You just pop in different modules for these different categories of metrics,
00:02:56.24 which include accounting how many of a thing exist;
00:02:59.02 sizes; shapes;
00:03:00.22 textures, which is the smoothness of a fluorescent intensity staining pattern;
00:03:04.22 and as well as the amount of intensity, which can correspond to the actual amount of a protein product, for example, in your images;
00:03:10.12 as well as things like spatial relationships.
00:03:13.13 Now, once your pipeline is set up to your liking,
00:03:15.15 you can run it on many images automatically.
00:03:17.25 Now, if it's a small number,
00:03:19.19 you might run it on your laptop or desktop.
00:03:21.07 If it's a very large experiment,
00:03:23.01 you might need to run your images on a computing cluster
00:03:25.10 or use cloud resources online.
00:03:27.15 And there's tools to help you do both.
00:03:29.21 Finally, you can explore your data using, really,
00:03:32.18 any data analysis software.
00:03:34.07 One option is CellProfiler Analyst.
00:03:36.11 It is designed to allow you to explore data from large sets of images
00:03:38.23 where the data is interactively linked to the images.
00:03:41.16 It also lets you classify phenotypes automatically.
00:03:43.22 Let's look at both of these kinds of features.
00:03:45.28 First, the exploration tools.
00:03:47.23 It contains many data visualizations
00:03:49.22 that you would see, really, in any spreadsheet software.
00:03:51.28 The difference is that in CellProfiler Analyst,
00:03:54.00 each datapoint is linked to the images that produced that datapoint.
00:03:57.08 This allows you to explore the features in your images
00:03:59.24 and the metrics that have been produced,
00:04:01.15 and try to identify what...
00:04:03.02 what's going on in your experiment,
00:04:04.20 or potentially even do some quality control measures
00:04:07.02 to identify if there's images that are blurry,
00:04:09.00 or have saturation or other kinds of artifacts in them.
00:04:12.04 And the most popular feature of CellProfiler Analyst is its classifier.
00:04:15.26 What it does is it displays a number of cells from your experiment
00:04:19.01 -- or other objects that you've identified --
00:04:21.08 and it asks you to sort them
00:04:23.21 based on their phenotype of interest.
00:04:24.29 And so you...
00:04:26.15 all you have to do is just drag and drop individual cells
00:04:28.11 to sort them as positive and negative for a phenotype,
00:04:31.05 or really you can have as many bins as you like.
00:04:32.25 And then as you are sorting the individual cells,
00:04:36.03 the computer is learning from you and trying to identify,
00:04:38.24 what are the metrics of the cells that distinguish the different bins?
00:04:42.03 As you are at sorting more and more cells,
00:04:45.02 and as you're correcting errors,
00:04:46.26 the classifier gets better and better,
00:04:48.14 to the point where the computer can replace you in making the correct answer
00:04:52.20 for a large number of cells.
00:04:54.10 So, once the classifier is trained,
00:04:56.01 you can sort... you can score millions or billions of cells
00:04:59.13 in a completely automated fashion.
00:05:00.24 My lab has been working with several others around the world
00:05:03.10 to create a new web-based tool for classifying images
00:05:05.24 that's powered by deep learning.
00:05:07.16 So, take a look on its website
00:05:09.08 to see if that's ready for use.
00:05:11.26 You can learn more through these related iBiology videos
00:05:14.18 on microscopy and image analysis.
00:05:16.10 And I hope you'll give bioimage analysis a try,
00:05:18.19 using CellProfiler,
00:05:20.04 and head to the online Scientific Community Image Forum if you need help getting started.

This Talk

Speaker: Anne Carpenter

Audience:

Educators of Adv. Undergrad / Grad
Researcher
Educators

Recorded: April 2019

Talk Overview

In this talk, Dr. Anne Carpenter provides an overview of CellProfiler, a free, open-source software program for image analysis. CellProfiler helps scientists to identify and measure biological entities, process images, and export data for further analysis. Carpenter provides examples on how to use CellProfiler, and explains how CellProfiler can aid scientists in their bioimage analysis.

Speaker Bio

Anne Carpenter

Anne Carpenter is an Institute Scientist at the Broad Institute of MIT and Harvard. Carpenter completed her bachelor’s degree in biological sciences, and a doctoral degree in cell biology from the University of Illinois, Urbana-Champaign. She continued her scientific training as a postdoctoral fellow at the MIT/Whitehead Institute for Biomedical Research in the laboratory of… Continue Reading