Reproduce the results for "Turning single-cell perturbation models into target-directed drug-combination screens"

These are datasets used to run PHAROS: a command-line application for predicting drug combinations that move cells from a user-defined starting state toward a target state. Specifically, they are used to generate the key reference databases that every dataset is queried against. This dataset also contains a pre-processed and embedded example ready for PHAROS predictions and used in the tutorial. tahoe100m_stse_manifold_reference.tar.gz is a parquet vector database. This is used to detect if a query dataset falls within the learned manifold of the STATE single-cell perturbation model ST-SE. This contains the cell lines from Tahoe-100M databse: all 50 cell lines consisting of WT cells and cells with 5uM drug perturbation across all perturbations. Tahoe100m_3_cell_lines_alldrugs_5uMconc_log1p_norm10k.SE600M.h5ad is a subset of the Tahoe-100M dataset. These are 3 randomly sampled cell lines with all 5uM drug perturbations with 300 cells for each cell-drug state. This dataset has been processed and embedded using the STATE single-cell foundation model SE600M. Tahoe100m_allWT_cell_lines_alldrugs_5uMconc_log1p_norm10k.SE600M.h5ad is a subset of the Tahoe-100M dataset. These are all WT (DMSO) cell lines with all 5uM drug perturbations with 100 cells for each cell-drug state. This dataset has been processed and embedded using the STATE single-cell foundation model SE600M. GSE206741_qc_mad_scrublet_log1p.pano_criz.SE600M.h5ad is one positive control example dataset that has been processed and ready for PHAROS conversion prediction. This is taken from GEO GSE206741. It contains A549 cell line with DMSO starting state and a target state perturbed with panobinostat + crizotinib. This dataset has been processed and embedded using the STATE single-cell foundation model SE600M. Please visit https://github.com/jbezney61/PHAROS and https://github.com/jbezney61/PHAROS_reproduce to see more detailed implementations and use cases of these datasets.