Skip to main content

Advertisement

Fig. 1 | Journal of Cheminformatics

Fig. 1

From: A multi-label approach to target prediction taking ligand promiscuity into account

Fig. 1

Different classification schemes. (a) Single-label binary classification scheme. Purinergic receptor P2Y12 (ChEMBL2001) shown in cyan and Butyrylcholinesterase (ChEMBL1914) shown in purple illustrate a binary classification problem. In binary classification, only 2 disjoint classes exist. Therefore in this classification scheme, |L| = 2 and |Y| = 1. (b) Single-label multi-class classification scheme. Phosphodiesterase 10A (ChEMBL4409) in green has been added to a single-label binary classification problem to form a single-label multi-class classification problem. That is to say, a single-label multi-class classification has more than two disjoint classes; hence, |L| > 2 (in this case |L| = 3) while |Y| = 1. (c) Multi-label multi-class classification scheme. Serine/threonine-protein kinase PIM1 (ChEMBL2147) in yellow, Protein kinase C delta (ChEMBL2996) in magenta and c-Jun N-terminal kinase 1 (ChEMBL2276) showed in grey, illustrate a multi-label multi-class classification. In multi-label classification problem, classes are not disjoint and compounds can belong to more than 1 class. Here, compounds shown in red belong to all 3 classes and have |Y| > 1 (in this case |Y| = 3). Furthermore, like the single-label multi-class classification problems, in this multi-label multi-classification scheme the value of |L| is bigger than 2 as we are dealing with more than two classes.

Back to article page