Fig. 7

PTM-wise common words in train, test, large scale predictions. The high quality predictions from the large scale extraction have picked up “key terms” associated with the interaction type, e.g. phosphorylation predictions have commons words such as phosphorylation and kinase. The low quality predictions from the large scale extraction on the other hand have generic words such as cell, activity and expression. For ubiquitination, the model seems to not have picked up the “key terms” across test and large scale predictions, and even in the training set, the term “ubiquitination” has relatively low representation compared to terms such as “protein(s)” and “cell(s)”