Local Function Conservation in Sequence and Structure Space

We assess the variability of protein function in protein sequence and structure space. Various regions in this space exhibit considerable difference in the local conservation of molecular function. We analyze and capture local function conservation by means of logistic curves. Based on this analysis, we propose a method for predicting molecular function of a query protein with known structure but unknown function. The prediction method is rigorously assessed and compared with a previously published function predictor. Furthermore, we apply the method to 500 functionally unannotated PDB structures and discuss selected examples. The proposed approach provides a simple yet consistent statistical model for the complex relations between protein sequence, structure, and function. The GOdot method is available online (http://godot.bioinf.mpi-inf.mpg.de).Author Summary: Proteins are an essential class of molecules playing a variety of roles within a cell. They can be described in various ways: amongst others, by sequence, structure, and function. Determining protein function by wet lab procedures is challenging and tedious. Simultaneously, sequencing and structural genomics projects turn out ever increasing numbers of protein sequences and structures, which are largely lacking functional characterization. As a consequence, there is a growing demand for computational methods that can assist human experts in the functional annotation of proteins. We present a method for protein function prediction based on a novel concept, called local function conservation. Local function conservation in sequence and structure is determined by rigorously analyzing the variability of protein function with respect to sequence and structure similarity. Our method predicts protein function even if the protein to be functionally annotated has only distant relatives. Furthermore, we estimate the reliability of the function prediction. With this approach, we advance automated function prediction and contribute to a better understanding of the complex relations between protein sequence, structure, and function.