Directory contains a zip file comprised of two CellProfiler pipelines (one per experimental replicate of the test dataset) for segmenting images, tracking objects between frames, and generating measurements of size, shape, and movement, in addition to the rules for user-defined state classifications (Round, Spread, Spindle). The rules for user-defined state classification are provided as a text file, which may be loaded into the FilterObjects modules of the pipelines. This directory also contains a PDF file showing the settings used to analyse the sample datasets, and a template experiment key.
Directory contains KNIME pipeline and associated user manual (as well as Conda environment) for software setup, running instructions, tips, and some troubleshooting information. In addition, a subdirectory called "Demo Videos" contains videos demonstrating data import and analysis.
Expected installation and set-up time for all dependencies is approximately 2 hours. The KNIME pipeline was tested on a machine running Windows 10, with 64GB RAM and an Intel Core i7-7740X CPU @ 4.30GHz processor. Expected runtime of the sample data on a machine with similar CPU specifications is approximately 4 hours for all components of the pipeline. Most users will typically only run either the user-defined or data-driven sections, not both.
Sample data (ExampleshRNA_live cysts.zip) and expected output (ExpectedOutput.zip) can be obtained from the following URL: https://www.dropbox.com/sh/igfpnt174l14ftc/AADq5B6RapGfJ16VOjTawUq9a?dl=0
After KNIME (with R and Python integrations) set-up, open the provided pipeline (Traject3D_v1.knwf) in KNIME. Uncompress ExampleshRNA_live cysts.zip and load the resulting "ExampleshRNA_live cysts" directory into the first node/component of the pipeline. This is a dummy dataset containing a scrambled shRNA control sample ("Scramble"), and two shRNAs ("shRNA_1", "shRNA_2"). Two experimental replicates were performed, each with four wells per condition, which were imaged hourly for 2 days (48hrs). The images were segmented, and acini tracked using CellProfiler. A user-defined shape classification ("Round", "Spindle", "Spread") was also applied. We provide the output from this CellProfiler analysis.
Due to the variability in tSNE and PhenoGraph analyses, even when using the same parameter values the output will not be exactly identical, but it should be similar.
Settings used to generate expected output:
- In "Select Samples and Frames to Include" (Data pre-processing section) component, "filter by timepoints" was set to "no" in order to include all timepoints in the subsequent analysis.
- In all components where time chunking is applied ("Calculate TimeChunks", "Trajectory Timechunk Plots (Frequency Motifs and Transitions)"), we used a time chunk of 6hrs.
- User-defined classification: For identification of user-defined classification trajectories, PhenoGraph k-nearest neighbours = 50.
- Data-driven classification: For identification of data-driven states, data was subsampled to 20,000 points using GeoSketch. PhenoGraph k-nearest neighbours = 40. For identification of trajectories, PhenoGraph k-nearest neighbours = 40. Data was subsampled to 20,000 points using GeoSketch prior to tSNE. tSNE was then performed with PCA initiation, iterations = 5000, theta = 0.25, perplexity = 50.
To get started, we provide an extensive user manual, which includes a step-by-step guide, videos demonstrating Traject3d in action, and a question section about what may be common errors. We will endeavour to provide genuine assistance to those that have first obtained local support from their IT to install Traject3d if needed. In the case of needing to do something that we have not tested for, this would need to occur as a discussion and potential collaboration.
We have setup a dedicated query email (address available on Traject3d GitHub account) for this: query.traject3d@gmail.com
In addition, we also expect that interested parties might directly email the corresponding author (david.bryant@glasgow.ac.uk) if they required assistance.