Extended Data Fig. 3: Performance of spatial clustering methods across different parameters. | Nature Cancer

Extended Data Fig. 3: Performance of spatial clustering methods across different parameters.

From: The automated computational workflow QUICHE reveals structural definitions of antitumor responses in triple-negative breast cancer

Extended Data Fig. 3

(a-b) Heatmap shows the performance of spatial clustering methods, K-Means++ and CellCharter, on detecting differentially-enriched cellular niches on (a) unstructured spatial topologies and (b) structured spatial topologies across different clustering resolution parameters. (c-e) Multi-patient spatial proteomic data were simulated with structured spatial topologies (that is, niche size of 50 immune cells within 500 pixel radius and niche prevalence of 100% samples). QUICHE achieves similar (c) niche recall and (d) niche purity scores across a range of hyperparameters, including the number of nearest neighbors in spatial niche detection, the radius size in spatial niche detection, the number of niches sampled from each patient sample, and the number of k-nearest neighbors in niche similarity graph construction. (e) Runtime performance of QUICHE as you vary (left) the number of niches selected from each patient sample in a cohort with 100 samples or (right) the cohort size, where each patient sample has 1000 niches. The error bands show the mean +/- standard deviation over n=5 random trials. Source data is provided in the Source Data file.

Back to article page