GENOME-3D: a UK network providing structure-based annotations for genotype to phenotype studies

点击翻译按钮获取中文摘要

The 3D structures of proteins are essential to fully characterise the sites mediating their molecular functions and their interactions with other proteins. However, whilst revolutionary technologies have enabled the sequencing of thousands of complete genomes, it is more challenging to determine the 3D structures of the proteins. Although the sequence repositories now contain >10 million protein sequences, less than 70,000 protein structures have been determined. Fortunately, in parallel with developments in sequencing technologies, powerful computational methods have emerged to predict the structure of a protein from its sequence. Currently these methods provide putative structures for ~80% of domain sequences from completed genomes, although the accuracy of this data varies from reasonably precise when structures are modelled using templates based on close relatives, through to quite approximate for models based on remote relatives and where proteins have no structurally characterised relatives. This project will bring together 6 internationally renowned UK groups involved in (1) classifying protein domains into evolutionary families (as this facilitates structure and function prediction) and/or (2) protein structure prediction. As regards the first activity - classification of protein structures - the two groups involved (SCOP,CATH) are the only groups, worldwide, providing this data. However, each applies somewhat different methodologies to make their assignments. Collaboration between these groups, in GENOME-3D, will involve comparison of domain structures and family classifications leading to refinements of assignments and/or confidence levels where the methods disagree. Since manual curation of the data is essential and since the rate at which the structures are determined is increasing, collaborations will speed up classification by allowing the groups to share information on the more challenging assignments and to discuss outcomes. For the second activity, structure prediction, the groups involved use technologies that vary in their sensitivity and in their ability to handle large numbers of sequences. Whilst SUPERFAMILY (based on SCOP) and Gene3D (based on CATH) provide greater coverage they are less likely to recognise very remote homologues, where methods such as GenTHREADER, Phyre, Fugue perform better. For each sequence, we will combine predictions from these different resources and assign confidence for each residue position in a query sequence based on the number of methods that agree in their structural prediction. We will provide pre-calculated assignments and also allow dynamic queries on the methods. We will also build 3D models for the sequences with residue positions highlighted according to agreement between the methods. We will develop computational platforms that integrate the information provided by each resource. To distribute this data to the biological and medical community we will build a dedicated web site. We will also establish web servers that link the methods ie run all the methods on query sequences and then report consensus assignments and highlight differences. In addition the consensus classification and annotation data will also be provided via two major international sites - the PDBe and InterPro. The sequence repositories are expanding at phenomenal rates as metagenomics and next gen sequencing initiatives bring in sequences from diverse microbial environments and report sequence variants occurring across different human populations or associated with different disease phenotypes. Structural data will enhance the insights available from this data. For example, known or predicted structures can reveal whether residue mutations oc