PMC

Procedure: clustering on confusionmatrix (ConfusionMatrix, threshold).

Flow chart for clustering protein subcellular location patterns.

Histograms of selected features before z-score normalization. Examples of features with (a) a roughly Gaussian distribution (3D-SLF11.6, average object to center of fluorescence distance), (b) a roughly Poisson distribution (3D-SLF11.23, texture feature average of co-occurrence matrix sum variance), and (c) a biomodal distribution (3D-SLF11.37, texture feature range of co-occurrence matrix sum entropy).

Selected images from the 3D 3T3 image dataset. Tagged protein names are shown with a hyphen followed by a clone number if the same protein was tagged in more than one clone in the dataset. Representative images are shown for (a) Atp5a1-1, (b) Ewsh, (c) Glut1, (d) Tubb2-1, (e) Canx, and (f) Hmga1-1. The top portion of each panel shows a projection on the x-y plane and the bottom shows a projection on the x-z plane.

A consensus subcellular location tree generated on the 3D 3T3 image dataset using SDA-selected 3D-SLF11 features. The columns show the protein names (if known), human observations of subcellular location, and subcellular location inferred from gene ontology (GO) annotations. The sum of the lengths of horizontal edges connecting two proteins represents the distance between them in the feature space. Proteins for which the location described by human observation differs significantly from that inferred from GO annotations are marked (**).