G Protein-coupled receptors (GPCRs) are integral cell membrane proteins of great relevance for pharmacology due to their role in transducing extracellular signals. The 3-D s tructure is unknown for most of them, and the investigation of their structure-function relationships usually relies on the construction of 3-D receptor models from amino acid sequence alignment onto those receptors of known structure. Sequence alignment risks the loss of relevant information. Different approaches have attempted the analysis of alignment-free sequences on the basis of amino acid physicochemical properties. In this paper, we use the Auto-Cross Covariance method and compare it to an amino acid composition representation. Novel semi-supervised manifold learning methods are then used to classify the several members of class C GPCRs on the basis of the transformed data. This approach is relevant because protein sequences are not always labeled and methods that provide robust classification for a limited amount of labels are required.

Peer Reviewed