Add PCA (dimensionality reduction) compute and plot

[asizemore]

For the differential expression notebook.

This PR works with PR 75 in service-eda. Can use qa mbio to test and see example requests in that PR.

Tasks

• Create computation
• Create visualization

Notes:
Can't use the scatterplot straight up because the dim red plot needs to set its own x/y axes with the computed variable (in beta div, the client doesn't set x/y. it just gets only that stuff from the backend). So I used the Options feature to have the new compute dimensionalityReduction tell ScatterplotVisualization that it wants to use these computed variables and that these computed vars need to be used in the request to the backend. We could have done it more simply with some hard coding but this will make it easier for us to add the ability to choose which principle coordinate the user wants for each axis. The backend is already prepped for this new feature, so the frontend might as well be, too.

Testing (updated 11/04/25)

• Backend merged on 11/03 so please use with ablevins mbio dev site backend with the local eda workspace frontend (yarn ... /eda).

IMPORTANT - do not merge with index.ts still including this compute. The backend has mbio approved for this compute because we need it for data testing. We can merge all this work but we must take it out of index.ts first so that it doesn't accidentally show up on the mbio live site someday on accident 😬

[Copilot]

Pull Request Overview

This PR adds a new dimensionality reduction computation plugin and enhances the scatterplot visualization to support computed variables in backend requests. The implementation follows the pattern of similar plugins like beta diversity.

• Adds a dimensionalityReduction computation plugin with PCA axis support
• Introduces sendComputedVariablesInRequest option to control whether computed variable descriptors are sent to the backend
• Registers the new plugin in the computation plugins index

Reviewed Changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated 6 comments.

File	Description
packages/libs/eda/src/lib/core/components/visualizations/implementations/ScatterplotVisualization.tsx	Adds sendComputedVariablesInRequest option and uses it to conditionally send computed X/Y axis descriptors to backend
packages/libs/eda/src/lib/core/components/computations/plugins/index.ts	Registers the new dimensionalityreduction plugin
packages/libs/eda/src/lib/core/components/computations/plugins/dimensionalityReduction.tsx	New computation plugin for dimensionality reduction (PCA) with configuration UI and computed axis details

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[asizemore]

More testing notes - the PR as it is can only be tested 1. with mbio and 2. in the regular eda workspace (not the notebook).

[asizemore]

Tagging @bobular

[bobular]

This looks good. A few minor comments we can discuss if needed. I assume it's all working, but I haven't tested.

Is there another PR where this is integrated into a notebook?

[bobular]

So we just "know" that "PC1" is a valid variableId here? This is a computed (not curated) variable collection, so I guess "PC%d" is defined on the back end somewhere. It's not going to change, I guess. Just wanted to understand better. I'm not demanding hugely complex changes to make this more bulletproof than it needs to be!

[asizemore]

Yes, for right now we just "know", and yep you're right it's defined in the backend. The differential expression function in veupathutils returns multiple PCs and you can even tell it how many to return. I had started to implement support for choosing PCs but then left it alone because it felt like it was making the initial deliverable too big.

[bobular]

Line 382 (variablesForConstraints) should probably be calculated consistently too? (including sendComputedVariablesInRequest in the conditional?

e.g.

  xAxisVariable: sendComputedVariablesInRequest && computedXAxisDescriptor
    ? computedXAxisDescriptor
    : vizConfig.xAxisVariable,