Clumppling 2.0: A Clustering Alignment Program for Population Structure Analyses

We previously introduced Clumppling to address the "alignment problem" for multiple mixed-membership unsupervised clustering results in population structure analyses, where clusters represent latent genetic ancestries. This problem stems from three challenges-label-switching, multi-modality, and varying numbers of clusters-which Clumppling resolves in three steps: aligning results with the same number of clusters, detecting distinct solutions or "modes," and aligning modes across different numbers of clusters. Here, we present Clumppling 2.0, an update with features for visualizing the emergence of clusters, comparing aligned results from different models, and incorporating modularity of algorithmic steps. We outline the Clumppling 2.0 workflow, highlighting its improved algorithmic flexibility and visual interpretability through a graph of alignment patterns. We then demonstrate its utility on human genetic datasets that include individuals from admixed populations.