Consensus protein localization encodings for all OpenCell targets

This dataset consists of the raw consensus encodings (or latent-space representations) of the protein localization patterns for all targets in the OpenCell library. These encodings are generated by a self-supervised machine learning model called cytoself trained on the OpenCell imaging dataset. Briefly, these encodings correspond to the flattened VQ2 layer of the VQ-VAE2-like component of the cytoself model, and are provided here as a mean over all images of each target. The resulting matrix of encodings has dimensions of 1294x9216, corresponding to the library of 1294 OpenCell targets and the 9216 dimensions of the flattened VQ2 layer of the cytoself model. This matrix is provided along with target metadata in the form of an `anndata` object. Please refer to our GitHub repo (2021-opencell-figures) for usage examples.