Fig. 4: Expert human evaluation and the confidence score of paragraphs generated by GO2Sum.
From: GO2Sum: generating human-readable functional summary of proteins from GO terms
a Pearson correlation coefficient between pairs of human evaluators for the dataset of 100 evenly distributed entries. b Comparison between the average embedding score and the average of human evaluators score. The score correlation for the 100 proteins entry dataset has an average embedding score between 0.1 to 1.0. c the second dataset of 45 protein entries with a high average embedding score between 0.8 to 1.0. d Distribution of the confidence score for entries classified into three classes based on the average embedding score. High, [0.8, 1.0]; moderate, [0.5, 0.8); low, [0, 0.5). Function paragraphs. 4591, 4400, and 769 proteins were included in the high, moderate, and low score class, respectively. e Subunit Structure paragraphs. 3046, 2806, and 382 in high, moderate, and low. f Pathway paragraphs. 1329, 121, and 10 proteins in high, moderate, and low.
