Additional file 1 of Zero-shot evaluation reveals limitations of single-cell foundation models

Additional file 1: Supplementary Tables S1-S7. Table S1: Scores for cell embeddings generated by the models. Scores calculated based on the whole dataset, as well in 10 repeats of subsampled for 10% of the dataset. Table S2: Rankings of median scores across datasets. Rankings are based on the median scores obtained from the evaluated datasets. A lower rank indicates better performance, highlighted in blue, while a higher rank signifies poorer performance, highlighted in red. Table S3: Source of the datasets and overlap with the data used for pretraining of the models. Table S4: Average ranking for the evaluated models. Rankings were calculated based on mean scores for biological integrationas well as batch mixing. A lower rank indicates better performance, highlighted in blue, while a higher rank signifies poorer performance, highlighted in red. Table S5: Scores for UMAP of the cell embeddings generated by the models. Scores calculated based on the whole dataset, as well in 10 repeats of subsampled for 10% of the dataset. Table S6: Average ranking for the UMAP embeddings of the embeddings generated by the evaluated models. Rankings were calculated based on mean scores for biological integrationas well as batch mixing. A lower rank indicates better performance, highlighted in blue, while a higher rank signifies poorer performance, highlighted in red.mean scores for biological integrationas well as batch mixing. A lower rank indicates better performance, highlighted in blue, while a higher rank signifies poorer performance, highlighted in red. Table S7: Overview of the used datasets.