Extended Data Fig. 5: cytoself discriminates between lysosomal and endosomal proteins.
From: Self-supervised deep learning encodes high-resolution features of protein subcellular localization
(a) We selected 11 proteins in OpenCell annotated in Uniprot as lysosomal or endosomal that are independently confirmed as such by mass spectrometry41,42. We show that cytoself is able to distinguish the lysosomal from endosomal proteins solely on the basis of the fluorescence images. All of these proteins are annotated on the basis of the images as ‘vesicles’ in both HPA and OpenCell. The min and max intensities of each image are adjusted to ensure comparable visibility. All representative images were randomly selected from each protein. Scale bar: 10 μm. (b) Clustering of these proteins on the basis of the feature spectra. (c) Feature spectra correlations for pairs of lysosomal and endosomal proteins, and for mixed lysosomal-endosomal pairs. Each box indicates the extent from the first to the third quartile of the data, with a line representing the median. The whiskers indicates 1.5 times the inter-quartile range. The p-values are computed using two-sided Mann–Whitney U test.
