Comparative embedding visualization for single-cell data

Do you frequently explore single-cell data via 2D embedding visualizations?

Comparing multiple embeddings can be challenging due to batch effects, missing point correspondences, and the sheer complexity of of the data.

I’m super excited for Ozette Technologies’s first graduate research intern Trevor Manz to present his research work on addressing several of these challenges in comparative embedding visualization at Cyto’s “High Dimensional Cytometry” session today (Room 510B at 10:30am).

➡️ https://github.com/OzetteTech/comparative-embedding-visualization

Following our work on annotation-transformation embedding approach (https://github.com/flekschas-ozette/ismb-biovis-2022), Trevor developed graph-based metrics for comparing embeddings via shared point labels. The metrics quantify how much point labels are visually intermixed, how much the local neighborhood of labels differ between two embeddings, and how much the (cluster) size of labels change within their local neighborhoods between two embeddings.

Moreover, all three metrics can be applied across label hierarchies and to different samples as long as the labels are shared between the embeddings.

We developed a Python-based prototype for the Jupyter Lab/Notebook environment that you can find at https://github.com/OzetteTech/comparative-embedding-visualization. The API isn’t stable yet but feel free to give it a try! And let us know what you think. 🙏

If you miss Trevor’s talk, feel free to drop by our booth (Number 539) to learn more about our ongoing work and Ozette Technologies new suite of ML-powered tools for analyzing and exploring high-dimensional single-cell data!