Remote homology detection of proteins using 3D models enriched with physicochemical properties.

In this paper, a new method for remote protein homology detection called remote-3DP, is presented. The remote-3DP method is based on both predicted 3D information and physiochemical properties of amino acids. The remote-3DP method considers only 10 structural models to represent a protein and to distinguish between remote homologues and non-remote homologues in 54 SCOP families. The low dimensionality of the protein representation allows us to use different classification techniques and discover which one works better for each SCOP family. In this paper, we show that including a physicochemical property along with predicted 3D information into a local structural element, actually improves the accuracy in remote homology detection. The highest ROC score for a set of models that includes 3D information and the Hydropathy index reaches 0.953 on the SCOP 1.53 dataset. In addition, a model that ensembles the outputs of 10 physicochemical properties is built to make a consensus decision. The consensus strategy reaches a ROC score of 0.963 on the SCOP 1.53 dataset, surpassing the current methods based on sequence composition which accuracy range from 0.87 to 0.92. [ABSTRACT FROM AUTHOR]