For training of SuperParsing, FCNs, and DeepLab, we picked out 82 images with different weather conditions from the dataset and manually annotated them into four classes: bird, wind turbine, sky, and forest, which are all classes included in [22].
The bioinformatic analysis was combined into a single Excel table where the clusters were manually annotated and sorted into functionally related groups based on the BLAST results from the various databases.
Eventually they are manually annotated and placed into Swiss-Prot.
The ontologies with overlapping gene sets were manually annotated and grouped together into larger functional groupings (seen below in Figure 3 and Figure 4) that are summarized in Table S1 and Table S2.
The manually annotated queries were divided into two groups: queries with bibliographic information (such as Author Name, Journal Name, Title or other Citation information) and without.
Therefore, for each cell-type, the series/datasets of the manually annotated samples were partitioned into three sets with each set containing roughly the same number of samples.
The same bright-field channel was manually annotated to create manual ground truth ('mangt') for comparison.
We manually annotated a further 4727 proteins into non-chromatin and chromatin classes, the latter including functional subcategories ("Lab internal category" in Supplementary Table 1).
