JMP 17.2 > Complex Data Dimensions Reduction and Estimation > Methods

[#1]

Hi JMP Community,

I am working with complex gene expression data from different studies, for which I would like to estimate its intrinsic complexity. I came across the R package Rdimtools, which seems to provide the type of analyses I am interested in. Specifically, the end goal for these analyses is to define the optimal number of dimensions to capture the complexity of the data sets.

Are you aware of any methods in JMP (native or Add-Ins) that would address my needs?

Thank you.

Best regards,

TS

[#2]

Hi @Thierry_S,

I just briefly read the documentation about Rdimtools given the huge number of methods included. I think most common dimensionality reduction techniques are available in JMP/JMP Pro through the Multivariate platform:

• Principal Components : Linear dimensionality reduction technique based on correlation or covariances between variables. The data is linearly transformed onto a new coordinate system such that the directions (principal components) capture the largest variation in the data. 
• Multivariate Embedding : Non-linear dimensionality reduction technique, enabling to respect both local neighboring as well as global structure of the data. In this platform, you can either use UMAP (Uniform Manifold Approximation and Projection) or t-SNE methods.
• Factor Analysis : Linear dimensionality reduction technique that seeks to describe observed variables in terms of a smaller number of (unobservable) latent variables, or factors.

There are also some interesting use of other platforms that could help you :

• In Hierarchical Clustering, you can realize a Two Way Clustering, which enable you to cluster your rows (observations, like in any clustering technique), but also your columns (variables). In this platform, you can export distances matrix, which can be then used in other platforms, for example in Multidimensional Scaling.
• I never used Genetics tools, but since you mention gene expression as part of this work project, maybe some dedicated tools could be helpful.

Hope this first answer may help you,

[#3]

Don't forget Multidimensional Scaling (MDS) in JMP as well.

What do you plan to do with the knowledge of the complexity and reduced dimensions? Classification of gene groupings? Sample Groups? Prediction? Sorting of genes and/or samples to further interpret possible functional relationships?

Each method of dimension reduction has pros and cons. Some are better at certain types of structure in the data or are more helpful than others in understanding the kind of structure (or what is going to be done next with the data).