Free Data Analysis Tools
- olivertburton
- 2 days ago
- 4 min read
Flow cytometry experiments are getting larger and more complex. Especially with spectral flow, panels tend to be big nowadays, and that makes it harder to do conventional gating-style analysis efficiently. As a result, methods designed to deal with high parameter data such as dimensionality reduction and clustering have become more popular. While there are quite good fee-for-service platforms available for this, including OMIQ and Cytolution, today we’re going to look at a couple of the free options, which are rapidly improving.
In particular, this post covers:
If you're reading this and have written or know of another such tool, let me know.
CyCONDOR
What is it? cyCONDOR aims to be a complete ecosystem for analyzing your high parameter flow data. The idea here is to link together all the tools you might want to use in one easy-to-access environment where everything just works. So, instead of having to build yourself a custom-coded analysis pipeline, you can link elements in cyCONDOR.
How does it work? cyCONDOR is an R package. You install it from GitHub, either as a package or as a docker file, and interact with it in RStudio. A lot of the steps have been simplified for you by taking care of the details in the background, so the number of calls you have to make is relatively low.
What does it offer? You can do dimensionality reduction (tSNE, UMAP, diffusion maps, PCA), clustering (Phenograph and FlowSOM), pseudotime analysis, differential analysis with rigorous statistics, machine learning (for example, predicting clinical outcomes), and importantly, batch correction (Harmony, CytoNorm). There are some very nice tools for working with large data sets where you can pre-train a model on part of the data and apply it to the rest, which enables you to run the analysis much quicker. And, you can generate all sorts of publication-ready figures of your data.
cyCONDOR has excellent documentation and instructions online, and you can read the publication here.
MARMOT
What is it? MARMOT, first and foremost, is intended to be easy to use. You can do everything in a graphical interface (app), without have to type any code into a black box. That's a big plus.
What does it offer? Dimensionality reduction can be done via tSNE, UMAP or PaCMAP. Clustering options include FlowSOM, Phenograph and PARC. One of the best features of MARMOT is the ability to extensively customise the output graphs and plots without having to re-run any of the time-consuming analysis. This really enables you to make your figures the way you want them for publication.
CytoScribe
What is it? BioLegend's cloud-based flow cytometry analysis platform. Unlike the others, this is not open-source, so you don't get to see the workings behind it. There is a free tier, allowing you to hold up to 2GB (last I checked) of data on their servers. For more than that, you need to pay. Unlike MARMOT and cyCONDOR, CytoScribe is meant to be a full suite of flow cytometry tools, more like FlowJo or FCS Express. This means you can do all your gating and unmixing/compensation in the platform, which you cannot do in the others. It also means that if you want to change the selection of cells/events you used for running your analysis, you can update the gate and re-do it directly, whereas MARMOT and cyCONDOR rely on you doing gating and exporting FCS or CSV files from FlowJo or FCS Express.
What does it offer? Clustering (FlowSOM, not sure what else), dimensionality reduction (tSNE, UMAP), spectral unmixing and compensation, gating. Like in FlowJo or FCS Express, you can create gates on the dimensionality reduction plot, which can be handy. Everything is saved in your cloud account, so it's easy to return to an analysis and pick up where you left off. If you get the paid version, you can share your experiments with other people, which is quite nice for collaborative projects or even just checking how someone has set up their analysis if you're teaching.
CytoVI
What is it? CytoVI is a different beast. This is a Python-based variation autencoder framework for modelling and interpreting single-cell technology data. I'll revisit this after I dig into it a bit more because it is very complex. The basic idea, as I understand it, is that you can train a deep learning model on your data, and use that to make predictions and inferences about what's going on. As long as you include examples of all the types of data in your training set, you can integrate not only data from multiple batches of the same flow panel, but multiple overlapping panels or even cytometry with single cell RNA-sequencing.
What does it offer? Effective batch correction. Integration of multiple types of data or overlapping flow panels. If you have, for instance, one panel on surface markers and a second on cytokines or transcription factors, where both have the same gating markers, you can put the data together and impute (make statistical guesses) the likely expression pattern for the cytokines for the surface marker data. If you train your model once with cell type annotations (for example, you do your manual gating once and figure out what everything is), you can get cell type assignments out for any new data you put in. You can do differential expression analysis, and you can model/predict disease state or outcomes. One key feature that CytoVI provides, as an autoencoder, is probabilistic information about expression profiles, so you can know which markers a given cell expresses without using gates.
Comments