Processed Features In Support Of Liebeskind Et Al (2018)

Processed feature matrices used in Liebeskind et al. (2018). Supporting code: https://github.com/marcottelab/plum Datasets 1 - 4 correspond to those used in Figure 4: Dataset 1: No AP-MS, yeast CF-MS, training species: Human Dataset 2: AP-MS, yeast CF-MS, training species: Human Dataset 3: AP-MS, yeast CF-MS, training species: Human, Yeast Dataset 4: AP-MS, no yeast CF-MS, training species: Human, Yeast ".train_labeled.missing_annotated.csv" files are those used for training the model and include only orthogroups for which interactions are known in the training species. These known interactions come either from gold-standard test sets, such as CORUM or EMBL's training portal, or from the fact that at least of the orthogroup pairs is missing in the focal taxon. ".missing_annotated.csv" files were used for prediction, and include the entire feature matrices, plus known missing pairs. Note that there is no dataset 3 file. This is because data sets 2 and 3 differ only in the training species used, so dataset 3 predictions used dataset2_07302018.missing_annotated.csv as a feature matrix. dataset4_prediction_07302018.csv contains the predictions for all pairs on data set 4, the best performing data set that was used for all downstream analyses.