Comparative spatial clustering performance against mclust

[linjing-lab]

Hi, I compared Leiden algorithm based on graph deep learning with Leiden algorithm some time ago. It is obvious that the embedding representation based on graph is much better than Leiden algorithm, and mclust clustering algorithm performs well in many spatial transcriptomics methods. I want to know whether spatialleiden method and mclust method have made a comparison of HVG and PCA. SpatialLeiden shows the advantage of integrating spatial information when compared with non-spatial Leiden. However, under the premise of controlling the experimental conditions, I would like to know whether spatialleiden and mclust have advantages in spatial clustering with multimodal foundational model.

[niklasmueboe]

Thanks for the very interesting questions. We compared the performance of HVG, SVG and different PCA versions in Leiden and SpatialLeiden in our manuscript (see here). However, we did not specifically compare to mclust or compare to GNN models in detail as the objective of our study was to share SpatialLeiden with a community that already understands Leiden, thus making it easier to understand parameters. If you decide to look into SpatialLeiden vs mclust vs GNN (vs foundation models) in more details we would love to hear about your experiences.