Similarity

As described for the earlier tasks, the wildcards {participant_id} is a placeholder.

Configuration fields

data.references (optional)
parameters.clustering.validity.similarity parameters.clustering.n_clusters

Output

group/{n_clusters}clusters/individual_similarity.npy
group/group_similarity.tsv
group/cophenetic_correlation.tsv
group/reference_similarity.tsv (optional)

Benchmarking

benchmarks/k{n_clusters}.individual_similarity.log
benchmarks/group_similarity.log
benchmarks/reference_similarity.log (optional)

Individual-to-Individual Similarity

Using the individual participant cluster labels per cluster granularity k, a similarity matrix is computed containing the similarity scores between the cluster labels of each participant using the defined similarity metric. The sklearn package is used for all available metrics (sklearn.metrics) – adjusted Rand index, adjusted mutual information score, and the V measure score.

Individual-to-Group Similarity

Using the same individual participant cluster labels and the group-level cluster labels, the similarity is computed between the individual participant cluster labels and the group-level cluster labels per cluster granularity k. This is done using the same approach as the individual-to-individual similarity.

At this point the cophenetic correlation, computed in the previous task is merged.

Reference-to-Group Similarity

If reference images are provided, the labeled voxels are extracted from the images. The defined similarity metric is computed between each reference and each group-clustering result, done using the same apparoch as the individual-to-individual similarity.