The reduce dimensions function uses different techniques to reduce the dimensions of your data, which is particularly useful for the visualization of high-dimensional datasets.
Currently, we offer three different dimensionality reduction tools, including, Principal Components, t-Distributed Stochastic Neighbor Embedding (**t-SNE**) , and Uniform Manifold Approximation and Projection (**UMAP**). The different tools have different algorithm options and can each help visualize your data in a different way.
## How to Use
First, choose which tool to use and then change the *Algorithm Options* if you want to use something other than the default options.
Next choose which data to analyze and how to weight your data. You can include multiple samples, cell types, and channel mean fluorescent intensities (**MFI**). Additionally, you can either use the information from the neighborhoods (number of cells per neighborhood of each type, et.) or you can input your cells directly into the dimensionality reduction. If you use neighborhoods, you can visualize the different regions in your tissue, and if you use individual cells, you can visualize which cell types are present in your dataset. It is important to note here that although you can in principle include neighborhood position and volume information, this might cause errors and odd results. Additionally, if you include only a few very rare cell types the dimensionality reduction will not work very well.
After choosing which dataset to use you can select how your data is normalized as well as Normalize per Sample or Per Dataset.
Finally, just click OK and wait to see the resulting pictures. Once the algorithm is finished running, the resulting reduced dimension channels are added to your main data structure and can be viewed in any new figure you make.
## IsoMap
The full name of IsoMap is Isometric Mapping. IsoMap is one of the earliest approaches to manifold learning, which would be viewed as an extension of Multi-dimensional Scaling(MDS) or Kernel PCA. IsoMap seeks a lower dimensional embedding which maintains geodesic distances between all points.
The Isomap algorithm comprises three stages:
1.**Nearest neighbor search.**
2.**Shortest-path graph search.**
3.**Partial eigenvalue decomposition.**
#### Example
Below, we show an example of IsoMap in our application using the example dataset.
The first two pictures shows which options we used setting up the IsoMap.
The result of running IsoMap with these options is shown below. The running time depends on the performance of your computer, for a personal laptop it took around 1 minutes.
