Large-scale protein-protein post-translational modification extraction with distant supervision and confidence calibrated BioBERT

Table 9 Results of noise levels, after noise reduction, in training data to verify if the PPI relationship is described in the abstract

Interaction type	Correct	Not—sure	Total
Acetylation	4	1	5
Dephosphorylation	6	4	10
Deubiquitination	1	1	2
Methylation	4	6	10
Phosphorylation	6	4	10
Ubiquitination	2	3	5
Total	23	19	42

The training data is randomly sampled (10 samples per interaction type unless the number of available training samples is lower) and verified by a human annotator

ISSN: 1471-2105