6/6 Many thanks to @Pierre-Luc Germain, @markrobinsonca.bsky.social, @Ferdinand von Meyenn, the Robinson lab and the von Meyenn lab for their contributions and help!

💬 Questions, feedback, or ideas? We’d love to hear your thoughts! Drop them below or reach out.

5/6 All these metrics are implemented in poem 🧰🛠️, our new R package! It’s designed to make rigorous evaluation accessible and reproducible. github.com/RoseYuan/poem

4/6 Importantly, we also develop new metrics tailored to spatially-aware measurements, making them ideal for the growing field of spatial omics.

3/6 We propose a framework to systematically understand, compare, and select validation metrics for:
• Cell embeddings
• Graph construction
• Clustering
• Spatial domain detection

with the emphasis on 🌟biological relevance and 🌟bias awareness.

2/6 What makes a good metric? How do we interpret the results from different metrics? These questions are at the heart of our work. 🔍

1/6 Identifying subpopulations is central to single-cell analysis, but how do you know if your identification is good? External validation metrics help, but different metrics often give very different results. 🧐