Figure 2.

Conceptual visualization of our approach for incorporating functional inter-relationships into function predictions algorithms, and empirical evidence to show the abundance of proteins carrying labels similar to the target class in the neighborhood of a query protein (a) Overview of our approach. In addition to using the proteins annotated with the target class (dark red circles) in the query protein's neighborhood, our approach also uses evidence from proteins annotated with a class similar to the target class (lighter red circles) in order to determine whether the query protein belongs to the target class (b) Distribution of nearest neighbors of similar classes for a target functional class (GO:0006352) of medium size (54 proteins) in Mnaimneh et al's data set. The Y-axis denotes the semantic similarity of a functional class with the target class, and the X-axis shows the average number of proteins in that class that are included in a size 20 neighborhood of proteins annotated with the target class. Some of the classes (marked by arrows) have a significant similarity with the target class, and have an average frequency higher than that of the target class (similarity = 1) itself.