Putting a name on it
A chasm separates sequence data from the biology of organisms — and genome annotation will be the bridge, says Lincoln Stein, a bioinformatics expert at Cold Spring Harbor Laboratory in New York. Spanning three main categories — nucleotide sequence, protein sequence and biological process — annotation is the task of adding layers of analysis and interpretation to the raw sequences. The layers can be generated automatically by algorithms or meticulously built up by experts in the hands-on process of manual curation.
Because manual curation is time-consuming and genome projects are generating data, and even changing data, at an extraordinary pace, there is a strong motive to shift as much of the burden as possible to automated procedures. A major task in the annotation of genomes, especially large ones, is finding the genes. There are numerous gene-prediction algorithms that combine statistical information about gene features, such as splice sites, or compare stretches of genome sequence to previously identified coding sequences, or combine both approaches. A new type of algorithm, called a dual-genome predictor, uses data from two genomes, to locate genes by identifying regions of high similarity.
Each algorithm has its strengths and limitations, working better with certain genes and genomes than with others. The GENSCAN gene-predicting algorithm, developed by Chris Burge at the Massachusetts Institute of Technology, has become a workhorse for vertebrate annotation and was one of the algorithms used in the landmark publications of the draft human genome sequence. FGENESH, produced by software firm Softberry of Mount Kisco, New York, proved particularly useful for the Syngenta-led annotation of the rice genome sequence.
Good data preparation is also important. "A lot of the magic happens in the environment, not the algorithm," says Ewan Birney a bioinformatician at the European Bioinformatics Institute (EBI) in Hinxton, near Cambridge, UK. "People often focus on the whizzy technology to the detriment of the real smarts, which happen in the sanitization of data to present them to a hard-core algorithm." Data sanitization includes steps such as masking repetitive sequences, which can interfere with an algorithm's performance.
All current large-scale efforts involve a combination of automatic and manual approaches. "For me it's quite clear that they can only be complementary," says Rolf Apweiler at the EBI, who leads annotation for the major protein databases SWISS-PROT and TrEMBL. "You can't automate anything without having manual reference sets that you can rely on."
While Apweiler is tackling large-scale annotation, others are concentrating on finding genes and proteins linked to a particular process, such as a disease. The bioinformatics and drug-discovery company Inpharmatica in London, for example, provides annotation databases and tools to identify potential drug targets.
Because of the plethora of different names given to the same genes and proteins in different organisms, a growing trend is the use of 'ontologies' — controlled vocabularies in which descriptive terms (such as gene and protein names) and the relationships between them are consistently defined. One ontology that is now widely adopted is the Gene Ontology (GO), but it doesn't cover all biology, and others have developed their own, often complementary, ontologies. BioWisdom in Cambridge, UK, for example, sells information-retrieval and analysis tools for drug discovery based on proprietary ontologies in fields such as oncology and neuroscience.
Working as part of the Alliance for Cellular Signaling, a team led by Shankar Subramaniam is developing an ontology that captures the different states of a protein, such as phosphorylation state. This will serve as a foundation for the Molecule Pages, a literature-derived database of signalling molecules and their interactions.
GO coordinator Midori Harris at the EBI and her colleagues are encouraging developers of new ontologies to make them publicly available through GO's website. They hope this will not only drive standardization, but will help to expand GO's capabilities by allowing the creation of combinatorial terms derived from different ontologies.
But most researchers agree that tools are only part of the solution. "The passion for biology often gets missed out here," says Birney. "People think it is all about finding technical solutions that magically solve problems, but frankly, far more important is really wanting to see the data hang together."
No comments