Details about the methodology of visualization used in ESM Metagenomic Atlas

[OreFoX18]

Dear all,

I'm reaching out to inquire about the methodologies employed in the impressive visualization showcased in the ESM Atlas. We're currently working on a project where we aim to organize and cluster approximately 42,000 proteins, and we're eager to adopt similar techniques.

In particular, I'm curious about the process involved in generating the visualizations, especially regarding the utilization of UMAP. We've experimented with MMseqs2 to obtain clusters along with their respective representatives, followed by performing pairwise comparisons between these representatives. However, we're keen to understand the specifics of how UMAP was precisely implemented in your workflow.

Could you kindly provide insights into whether you also employed a similar approach of identifying cluster representatives and conducting pairwise comparisons to derive similarity scores for creating the distance matrix?

Your guidance and expertise would be greatly appreciated.

Warm regards,
Haris