Optimizations that are present in opt-SNE can be applied to the Fast Fourier Transform-accelerated Interpolation-based t-SNE (FIt-SNE). This will allow the processing of large datasets at a faster speed with the optimized parameters for cytometry data of opt-SNE.
opt-SNE is a variant of the t-SNE algorithm that features several improvements above the traditional Barnes-Hut implementation of t-SNE, including the ability to detect the rate of improvement of KL divergence (KLD - functionally, how good the low dimensional projection of many dimensions is) and then automatically stop the algorithm when it begins to suffer from diminishing returns in that metric. FIt-SNE, on the other hand, uses the Fast Fourier Transformation that computes the relationship of each data point to each other using an equispaced grid allowing for faster projections of high-dimensional data into a two dimensional space. Because FIt-SNE uses an equispaced grid large datasets can be handled more effectively.
You can read more about the two techniques below:
Belkina, A.C., Ciccolella, C.O., Anno, R. et al. Automated optimized parameters for T-distributed stochastic neighbor embedding improve visualization and analysis of large datasets. Nat Commun 10, 5415 (2019). https://doi.org/10.1038/s41467-019-13055-y
Linderman, G.C., Rachh, M., Hoskins, J.G. et al. Fast interpolation-based t-SNE for improved visualization of single-cell RNA-seq data. Nat Methods 16, 243–245 (2019). https://doi.org/10.1038/s41592-018-0308-4
1. Add a FIt-SNE task
Click Add new child task and select FIt-SNE from the task selector.
Note that your Workflow might look slightly different, if you are using Compact View.
2. Apply the Same Optimized Hyperparameter Strategy from opt-SNE to FIt-SNE
Shown above is the default settings for FIt-SNE in OMIQ. The settings that you would need to change in FIt-SNE to have the same hyperparameters of opt-SNE are Max Iterations; Stop Early Exaggeration; and Learning Rate.
2.1 Change the FIt-SNE settings to the following:
Max Iterations: 700
Stop Early Exaggeration: 150
Learning Rate: Compute for the Learning Rate using the formula: (# of events)/36.
For example, if you had 10 million events, the settings should be changed to the following:
Max Iterations: 700
Stop Early Exaggeration: 150
Learning Rate: 277,777 (10 million/36)