Protein structure prediction and structure-based protein function annotation

Abstract

Nature tends to modify rather than invent function of protein molecules, and the log of the modifications is encrypted in the gene sequence. Analysis of these modification events in evolutionarily related genes is important for assigning function to hypothetical genes and their products surging in databases, and to improve our understanding of the bioverse. However, random mutations occurring during evolution chisel the sequence to an extent that both decrypting these codes and identifying evolutionary relatives from sequence alone becomes difficult. Thankfully, even after many changes at the sequence level, the protein three-dimensional structures are often conserved and hence protein structural similarity usually provide more clues on evolution of functionally related proteins. In this dissertation, I study the design of three bioinformatics modules that form a new hierarchical approach for structure prediction and function annotation of proteins based on sequence-to-structure-to-function paradigm. First, we design an online platform for structure prediction of protein molecules using multiple threading alignments and iterative structural assembly simulations (I-TASSER). I review the components of this module and have added features that provide function annotation to the protein sequences and help to combine experimental and biological data for improving the structure modeling accuracy. The online service of the system has been supporting more than 20,000 biologists from over 100 countries. Next, we design a new comparative approach (COFACTOR) to identify the location of ligand binding sites on these modeled protein structures and spot the functional residue constellations using an innovative global-to-local structural alignment procedure and functional sites in known protein structures. Based on both large-scale benchmarking and blind tests (CASP), the method demonstrates significant advantages over the state-of-the- art methods of the field in recognizing ligand-binding residues for both metal and non- metal ligands. The major advantage of the method is the optimal combination of the local and global protein structural alignments, which helps to recognize functionally conserved structural motifs among proteins that have taken different evolutionary paths. We further extend the COFACTOR global-to-local approach to annotate the gene- ontology and enzyme classifications of protein molecules. Here, we added two new components to COFACTOR. First, we developed a new global structural match algorithm that allows performing better structural search. Second, a sensitive technique was proposed for constructing local 3D-signature motifs of template proteins that lack known functional sites, which allows us to perform query-template local structural similarity comparisons with all template proteins. A scoring scheme that combines the confidence score of structure prediction with global-local similarity score is used for assigning a confidence score to each of the predicted function. Large scale benchmarking shows that the predicted functions have remarkably improved precision and recall rates and also higher prediction coverage than the state-of-art sequence based methods. To explore the applicability of the method for real-world cases, we applied the method to a subset of ORFs from Chlamydia trachomatis and the functional annotations provided new testable hypothesis for improving the understanding of this phylogenetically distinct bacterium.