Skip to main content
Fig. 2 | BMC Genomics

Fig. 2

From: ACES: a machine learning toolbox for clustering analysis and visualization

Fig. 2

The workflow of ACES. ACES first reads the raw sample data file or distance matrix, and then automatically calculates the number and the potential centroids of clusters. Hierarchical clustering is set as the default, also allowing for k-means and DBSCAN. Initial input parameters are automatically estimated. To demonstrate the relationships among the samples together with the clustering results, the samples are downsized by PCA and visualized in 2D or 3D plots, colored by their cluster labels. Alternatively, the distance matrix is reordered for heat map visualization to show the clusters. ACES provides functionality to analyze data samples with multiple phenotypes to best explain the clustering: ACES automatically extracts and sorts all phenotypes/attributes and ranks them by consistency with the biomarker data, i.e. the discriminative power of each attribute is matched to the clusters in the data. The matches are then visualized in 2D/3D PCA plots as well as at the bottom of heat map, colored by attribute labels

Back to article page