Answers are grounded in this paper and cite their sources.
Recent full-genome sequencing projects have produced a wealth of DNA sequence data. This data, along with the anticipated completion of the human genome, has been lauded as the raw material of the biology of the next century. This is a bit misleading, however. It is not the linear lists of nucleotides or genes from each organism, but rather the list of associated biochemical functions, expression profiles and cellular roles that will provide this basic raw material. The storage, organization and retrieval of such information is the challenge being brought to bioinformatics. A genome informatics workshop was held in Washington 2–3 April 1998 by the United States Department of Energy and the National Institutes of Health to advise these funding agencies on the informatics needs to support and to exploit the various genome projects over the next five years. Among the many subjects discussed were: the need for the development of `Curated Reference Genome' data structures; `Gene function/role' data structures; tools to support `Comparative Genomics'; and tools to support `Regulatory Circuit Modeling'. While only the first of these was agreed on unanimously as the top priority item, they represent some of the major bioinformatics thrusts aimed at supporting the challenges facing functional genomics. William Gelbart (Harvard Univ., USA) and David Lipman (National Center for Biotechnology Information) noted that a four-way collaboration is planned between FlyBase[1FlyBase, http://flybase.bio.indiana. edu/Google Scholar], the two Drosophila Genome Projects (the Berkeley[2Berkeley Drosophila Genome Project, http://fruitfly.berkeley.edu/Google Scholar]and European[3European Drosophila Genome Project, http://edgp.ebi.ac.uk/Google Scholar]Projects), and the NCBI[4NCBI, http://www.ncbi.nlm.nih.gov/Google Scholar]to produce an expert-curated Drosophila melanogaster reference genome. The basic idea is that FlyBase (which now includes the informatics groups of the two associated genome projects) will provide continuously curated annotation. These annotations will be derived from the literature, comparative sequence analyses, and the FlyBase research community. NCBI will provide both access and GenBank[5GenBank, http://www.ncbi.nlm.nih. gov/Web/Genbank/index.htmlGoogle Scholar]archival linkages. Access to the reference sequence will be possible through other sequence databases, as well as through FlyBase. Ed Uberbacher (Oak Ridge National Laboratory) of the Genome Annotation Consortium[6Genome Annotation Consortium, http://compbio.ornl.gov/gac/index.shtmlGoogle Scholar]presented a pilot project for a similar effort on constructing a human reference genome along with new associated front end access computer tools. Here data ranging from EST projects to disease gene maps from OMIM[7OMIM, http://www.tigem.it/LOCAL/OMIM/info.htmlGoogle Scholar]can be included. Similar efforts are underway for Saccharomyces cerevisiae and a few of the other recently completed microbial genomes. The long-term aim of these reference genome projects is to create full-genome-sequence data structures to which synthesized and reviewed information can be attached. The information will be multilayered, ranging from single-base-position variation and mutations, through gene product biochemical function and expression levels, to genetic and phenotypic information associated with large regions. Perhaps as important as the direct linkage to such expanded annotation, will be the availability of the sources used by the curators for those annotations. These reference genomes will be searchable by sequence similarity and keywords as is now commonly done on GenBank and SWISS-PROT[8Bairoch A. Boeckmann B. Nucleic Acids Res. 1994; 22: 3578-3580Crossref PubMed Scopus (37) Google Scholar], but with considerably more retrievable information. In addition, they will be searchable in ways not currently possible. Gene families or homologues, genome regional context, cellular roles and pathways will all be queriable items, and the results of such queries will return access to vast amounts of associated, curated expert knowledge. Clearly, the availability of a dozen or so such reference genomes will change the way we do biology and the use and maintenance of the many current biological databases. With no vertebrate animal or higher plants yet sequenced, and with a large fraction of the encoded genes in the current genome sequences still having unidentified biochemical function[9Gray M.W. Nature. 1996; 383: 299Crossref PubMed Scopus (23) Google Scholar]or completely mysterious cellular roles; how close are we to a useful set of reference genomes? This is a bit of a `Catch 22' situation. We need considerably more detailed information about what is encoded in genetic sequences in order to fully annotate any reference genome. On the other hand, without some initial reference sequence structures, it is difficult to organize and collate our current knowledge or even to update it efficiently. With the anticipated increase in mRNA expression (EST), protein interaction and genetic regulation data—to say nothing of the wealth of population and phenotypic data—it was agreed that it is imperative that these reference structures be created now. We, as a community, probably have most of the tools and much of the data already in hand to create very useful starts toward this goal. It is relatively easy, for example, to identify even fairly distant gene relationships for which some functional information can be inferred. In addition, our ability to infer via sequence similarity function at the biochemical level will surely increase over the next couple of years. I assume that a database of diagnostic patterns and associated representative sequences for the majority of protein domains will soon become available. One can thus anticipate that each new gene will initially be dissectable into a set of biochemical functional domains by searching for matches through such a database. This is only the simplest of the challenges since one needs to determine the role, both temporal and spatial, that those biochemical functions play in the life of the cell and organism in question. This will be supported by having direct access to information as to what roles the domains play in other organisms and in what larger protein functional domain context they have previously been found. Clearly the identification of biochemical function and cellular roles across many genomes is a sort of boot-strapping process—the more we learn, the more we can hypothesize and/or extrapolate. Such testable hypotheses will drive much of experimental biology in the future. Why am I confident of this view? Over the last few years, various pattern representations, such as regular expressions, profiles, hidden Markov chains, neural nets and others have shown considerable success in recognizing the key sequence elements within a set of protein domains encoding a common biochemical function. While not yet as sensitive or complete as needed, such resources as the PROSITE[10Bairoch A. Nucleic Acids Res. 1992; 20: 2013-2018Crossref PubMed Scopus (423) Google Scholar], Pfam[11Sonnhammer E. Eddy S. Durbin R. Proteins. 1997; 28: 405-420Crossref PubMed Scopus (909) Google Scholar]and BLOCKS[12Henikoff S. Henikoff J.G. Nucleic Acids Res. 1991; 19: 6565-6572Crossref PubMed Scopus (378) Google Scholar, 13Pietrokovski S. Henikoff J.G. Henikoff S. Trends Genet. 1998; 14: 162-163Abstract Full Text Full Text PDF PubMed Scopus (9) Google Scholar]databases, and the iterative BLAST searches[14Neuwald A. et al.Nucleic Acids Res. 1997; 25: 1665-1677Crossref PubMed Scopus (202) Google Scholar], for example, have already proven very useful. These pattern tools have clear advantages over simply searching the ever-expanding archival sequence databases: first, a single match to a functional domain diagnostic pattern is a match to all other sequences containing that domain; second, the match can directly identify the key sequence elements and the linear extent of the domain essential for that function (Fig. 1). Thus, while much work is still needed on perfecting diagnostic pattern tools and protein domain dissection, one can clearly anticipate the day when, as with our childhood Lego or Erector toy sets, we will be less focused on the function of a protein than on how life's functional Legos are assembled to play their complex roles within living systems. Perhaps the most obvious problem in current efforts to identify protein function is the lack of controlled vocabulary and/or consistent nomenclature both in the literature and the databases. This is in part inherent in today's molecular biology with its many subdisciplines and rapid growth and changing knowledge base. The exception here may be in the Enzyme Commission nomenclature hierarchy, which unfortunately applies only to catalytic domains, and then only to nonmultiple subunit enzymes. These problems are compounded by the transitivity and dated nature of much of the current sequence inherited annotation in the current databases[15Doerks T. Bairoch A. Bork P. Trends Genet. 1998; 14: 248-250Abstract Full Text Full Text PDF PubMed Scopus (56) Google Scholar]. This is particularly true of multidomain and alternately spliced proteins. Here functional annotation is, at best, often properly associated with only one domain or splice variant, and in the worst cases has been inherited by a protein not containing the domain associated with the annotation[16Smith T.F. Zhang X. Nat. Biotechnol. 1997; 15: 1222-1223Crossref PubMed Scopus (48) Google Scholar]! While the majority of functions associated with known sequences were experimentally determined at some point, the vast majority of their recorded instances have been inferred only by comparative methods. Few biologists doubt the power of these comparative inferences as reflected in our current evolutionary understanding, for example. However, the backtracking of inherited annotation sources is often not possible without extensive literature searches given that the current databases do not generally provide information on how or why any particular annotation was assigned. Considerable collaborative curation efforts involving a wide range of expertise will thus be required to construct well-annotated databases of protein domain functions linked to the reference genomes. The existence of reference genomes will themselves help, via comparative genome analyses, to identify nomenclature inconsistencies and incompleteness as well as the range of cellular roles played by each distinct protein function. Comparative Genomics, while fuzzy in definition, will continue to be the central workhorse in molecular biology. The tools employed here include the standard sequence comparison and common pattern approaches, as well as whole-genome organization[17Koonin E. et al.Mol. Microbiol. 1997; 25: 619-637Crossref PubMed Scopus (266) Google Scholar]and statistical comparisons[18Freeman, J.M. et al. (1998) Science 279, 1827 www.sciencemag.org/cgi/content/full/279/5358/1827aGoogle Scholar]. One should probably think of the association of phenotypes with genetic variation as a form of comparative genomics. In fact, as we begin to try to understand the regulatory circuitry of the cell, one assumes that by comparisons of closely related organisms, we will learn much about how these circuits have been modified and augmented to carry out distinct developmental and other programs. In this last area of regulatory circuit modeling, we are just beginning. Here, as in the case of protein function, we do not yet possess a well-defined and agreed upon vocabulary. Researchers in different areas, such as embryonic development and central metabolism or among different species such as Saccharomyces cerevisiae and Drosophila melanogaster, use different gene names and functional and regulatory annotation. We have no real `algebra' for describing regulatory circuits across different systems. This is not to say that there are not attempts to formalize our description of some simple regulatory circuits. Considerable efforts have been made for the bacterial phages[19Ptashne, M. (1992) A Genetic Switch: Phage Lambda and Higher Organism (2nd edn) Cell Press/Blackwell Scientific PublicationsGoogle Scholar]and very recently for the regulation of the Endo-16 cis-regulated gene in the sea urchin[20Yuh C-H. Bolouri H. Davidson E.H. Science. 1998; 279: 1896-1902Crossref PubMed Scopus (560) Google Scholar]. The authors presented a modular logical-unit breakdown of this gene's promoter that is similar to the protein domain dissection problem. Given nature's duplication, rearrangement and modification strategy, we can anticipate that there may be some basic set of regulatory `Lego' pieces that can be assembled in a nearly infinite number of ways to create the developmental and behavioral diversity of terrestrial life. Simple sequence comparison or regulatory site pattern matches will not give much insight into the cell's regulatory circuits without considerable additional or associated information. This is not to say that studying the regulation of one gene at a time, and backtracking by hand all relative information in the literature has not or will not continue to be a fruitful approach. Rather, less and less of the vast amount of genome sequence, genetic variation and EST expression data will appear in the traditional literature. It will be placed instead directly in computer databases. In addition, more and more of this data's preliminary annotation will be automated. The efficient exploitation of such data will therefore also have to be partially automated. If the annotation errors and inconsistencies are not to overwhelm its value, we will need highly curated data structures organized to make genetic region-to-genetic region and organism-to-organism comparisons simple. Again, the creation of reference genomes with which we can associate diverse information will be a great help. The day does not seem far off when nearly all experimental biology will be driven by some kind of comparative genome suggested hypothesis. This will likely be the case not only in the evolutionary and developmental areas where it is already true, but in everything from ecology and population genetics to medicine. One has to assume that the current interest in function genomics by the pharmaceutical companies already demonstrates the latter. Caution is required on two fronts, however. First the lack of any holistic description of a cell's spatial and temporal interactions with its environment, let alone that of an entire complex organism, will require some kind of synthesis to sit above our current molecular-level reductionist views. Second, this long-term goal of the biological sciences will not be easily reached without major community-wide cooperative efforts in the curation and integration of the incredible wealth of data we will very shortly have in hand. I thank X. Zhang for the analysis used to construct Fig. 1. This work was supported by Grant P41 LM05205 from the National Library of Medicine and DE-FG02-98ER62558 from the US Department of Energy. Its contents are solely the responsibility of the author and do not necessarily represent the official views of the granting agencies.
No takes yet. Share an insight, caveat, or question.
Temple F. Smith (1998) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: