From: Exploiting likely-positive and unlabeled data to improve the identification of protein-protein interaction articles
Class
t i
¬t i
Positive
w
x
Negative
y
z