359
`
`Computational genetics: finding protein function by
`nonhomology methods
`Edward M Marcotte
`
`During the past year, computational methods have been
`developed that use the rapidly accumulating genomic data to
`discover protein function. The methods rely on properties shared
`by functionally related proteins other than sequence or structural
`similarity. Instead, these ‘nonhomology’ methods analyze
`patterns such as domain fusion, conserved gene position and
`gene co-inheritance and coexpression to identify protein—protein
`relationships. The methods can identify functions for proteins
`that are without characterized homologs and have been applied
`to genome-wide predictions of protein function.
`
`Addresses
`Molecular Biology Institute, UCLA-DOE Laboratory of Structural
`Biology and Molecular Medicine, University of California Los Angeles,
`PO Box 951570, Los Angeles, CA 90095-1570, USA and Protein
`Pathways Inc., 1145 Gayley Avenue, Ste 304, Los Angeles, CA
`90024, USA; e-mail: marcotte@mbi.ucla.edu
`
`Current Opinion in Structural Biology 2000, 10:359–365
`
`0959-440X/00/$ — see front matter
`© 2000 Elsevier Science Ltd. All rights reserved.
`
`Abbreviations
`COGS clusters of orthologs
`EST
`expressed sequence tag
`
`Introduction
`Biologists are in a delightful quandary. Thousands of
`potential genes are being discovered in the various
`genome sequencing projects, including those encoding
`many new families of proteins. Often, these proteins are
`evolutionarily conserved, but are of unknown function.
`This poses a fundamental problem to biologists: how can
`we discover the functions of these thousands of unknown
`proteins quickly and efficiently? Even more ambitious
`than knowing their specific biochemical functions, can we
`discover their broader functions — the cellular context,
`such as pathways and complexes, in which they operate?
`
`As difficult as this goal is, significant progress has been
`made in the past year both experimentally, by conducting
`genome-wide experiments measuring, for example,
`mRNA expression [1] or biochemical activity [2•], and
`computationally, by developing new analyses that work on
`fundamentally different principles from homology- or
`structure-based methods.
`
`This in silico progress stemmed from the realization that
`genomes contain considerable information about the func-
`tions of and relationships between genes and proteins.
`This functional information is encoded in forms such as
`patterns of gene fusion, conservation of gene position, pat-
`terns of gene co-inheritance and other sorts of
`evolutionary information. Such patterns are revealed by
`comparisons of multiple genomes, making these analyses
`
`only recently tractable. Also, additional data, such as gene
`coexpression measurements, provide analogous informa-
`tion within single organisms.
`
`The power of these new methods is that they produce net-
`works of functionally related proteins, even when the
`proteins have never been characterized. Protein function is
`defined by these methods in terms of context, that is,
`which cellular pathways or complexes the protein partici-
`pates in, rather than by suggesting a specific biochemical
`activity. However, in cases in which some of the proteins
`have a known function, their function can be extended to
`the most intimately linked uncharacterized proteins.
`Thus, the methods can be used both to find functional
`relationships and to assign general protein function.
`
`This results in an approach to finding protein function that
`is strikingly different from directly comparing amino acid
`sequences, although sequence comparisons are the basic
`tool used in many of the methods. The functional informa-
`tion discovered also differs from what might be learned
`either from direct sequence comparisons or from structural
`analyses, giving three relatively independent and comple-
`mentary routes to protein function, as shown in Figure 1.
`This review will discuss the main ideas behind nonhomol-
`ogy methods, the newest route to protein function.
`
`Evolution (some homology required)
`Several nonhomology methods take advantage of genetic
`variations among organisms to find protein function. The
`domain fusion method [3•] finds functionally related proteins
`by analyzing patterns of domain fusion. As illustrated in
`Figure 2, proteins found separately in one organism can
`often be found fused into a single polypeptide chain in
`another organism. That the separated proteins have a func-
`tional relationship can be inferred from knowledge of the
`fused protein, named the Rosetta stone protein for its abil-
`ity to reveal the relationship among its component parts.
`
`In many cases, the proteins linked by such a domain fusion
`event may even physically interact, especially in the case
`of protein pairs that have been filtered for false-positive-
`producing ‘promiscuous domains’ [3•] and in cases of
`high-scoring sequence matches to the Rosetta stone pro-
`tein [4•]. An example of this is the two Escherichia coli
`gyrase subunits GyrA and GyrB, which are found as fused
`homologs in yeast topoisomerase II [3•]. Such relationships
`are also common among separated eukaryotic proteins
`found fused in a prokaryotic Rosetta stone protein
`(EM Marcotte, unpublished data).
`
`In its simplest form, this analysis can be implemented
`[3•] by searching a large sequence database for homologs
`
`Petitioner Microsoft Corporation - Ex. 1037, p. 359
`
`

`

`360 Sequences and topology
`
`Figure 1
`
`Blast and
`Smith–Waterman
`
`Protein sequence
`
`Threading and
`fold recognition
`
`Figure 2
`
`(a)
`
`Homology
`
`Nonhomology
`
`Structure
`
`A
`
`B
`Organism 2
`
`D
`
`A
`
`D
`
`B
`
`F
`
`C
`
`Organism 3
`
`mRNA of C and D
`coexpressed
`in many different
`cellular conditions
`
`D
`
`Correlated mRNA
`expression
`
`C
`B
`Organism 1
`
`D
`
`A
`
`A
`
`Domain
`fusion
`
`Trees and clustering
`
`Domain fusions
`Phylogenetic profiles
`Correlated expression
`Conserved gene position
`
`Structural motifs and
`evolutionary tracing
`
`FUNCTION
`
`(b)
`
`Current Opinion in Structural Biology
`
`One can take several computational routes to discovering the function
`of a protein. On the left-hand route, the protein sequence is compared
`directly with other protein sequences [44,45]. Characterized sequence
`homologs or phylogenetic analyses (as in [17–19]) may suggest
`functional information. On the right-hand route, the protein sequence
`may be tested for compatibility with known three-dimensional protein
`structures [46]. Knowledge of the structure may then suggest
`functional information (e.g. as in [47,48]). Along the middle route are
`nonhomology methods. Sequence and structural homology reveal
`proteins of identical or equivalent function, whereas nonhomology
`methods identify interacting proteins, proteins with related functions or
`proteins operating in the same cellular context. Nonhomology methods
`return a network of relationships among proteins functionally linked to
`the query protein and function is both defined and inferred by this
`network of related proteins.
`
`of a query protein AA. Hits in this search will include
`direct homologs of the query protein (A′) and potential
`Rosetta stone fusion proteins (A–B). Each hit is then
`used as a query to search the genome of A and function-
`ally related B proteins will be found in this second
`search. Along with B proteins, hits in this search will
`include A, homologs of A (A′) and very distant homologs
`of A (A′′) [5], but the B proteins can be identified by
`their lack of homology to A or by their homology to dif-
`ferent regions of A–B than those homologous to A. Such
`an analysis recently proved useful in identifying a func-
`tional
`relationship between CHORD-containing
`proteins and Sgt1, proteins important for plant disease
`signaling and nematode development [6].
`
`In a related fashion, two proteins can be inferred to be
`functionally related if their genes are repeatedly found as
`neighbors on the chromosomes of different organ-
`isms [7•,8,9•], as shown in Figure 2. This conservation of
`relative gene position presumably derives from the organiza-
`tion of prokaryotic genes into operons in which each
`protein encoded by the operon performs a closely related
`task, such as the proteins of the lactose system [10] or pro-
`teins involved in iron uptake [11]. To find operons
`directly would require the identification of promoters and
`
`B
`
`Conserved
`gene
`position
`
`C
`
`Known metabolic
`pathway
`
`E
`
`Current Opinion in Structural Biology
`
`An example of deriving protein–protein relationships by nonhomology
`methods. Genes (labeled white boxes) are shown on the
`chromosomes (thick horizontal lines) of three different organisms. (a) It
`can be inferred that the proteins encoded by genes A–D are
`functionally related through patterns such as the conserved gene
`positions of B and C in organisms 1 and 3, the fusion of A and B into
`A–B in organism 2 and the coexpression of the mRNAs of C and D in
`organism 1. These results can be represented as a network of
`functional relationships, as shown in (b). If, for example, the function of
`B was unknown, it might be inferred from the functions of proteins A
`and C. The computational linkages may be supplemented by any
`experimentally observed interactions or known protein–protein
`relationships, such as those described in [35,38–40].
`
`regulatory elements; however, for operons containing evo-
`lutionarily conserved genes, large portions of the operons
`can often be reconstructed automatically simply by iden-
`tifying pairs of conserved gene neighbors [9•].
`
`As with domain fusion analysis, this approach can often
`identify interacting proteins [7•]. How well the approach
`will extend to eukaryotic genes remains to be seen, as
`eukaryotes generally lack operons. Examples of function-
`ally related eukaryotic gene neighbors do exist, however,
`such as in the TCL1 locus [12] or the cadherin pro-
`teins [13], so the technique may be useful. The quality of
`the functional relationships identified by this method is
`exceptional, but the coverage is unfortunately low because
`of the dual requirement of identifying orthologs in anoth-
`er genome and then finding those orthologs that are
`adjacent on the chromosome.
`
`Petitioner Microsoft Corporation - Ex. 1037, p. 360
`
`

`

`Nonhomology methods Marcotte 361
`
`A third nonhomology method works on the premise that
`proteins that operate together in the cell are often inherit-
`ed in a correlated fashion [14•]. That this is a reasonable
`assumption follows from the fact that proteins rarely work
`alone and many pathways or complexes are crippled by the
`loss of individual components. Thus, any organism that
`requires the complex or pathway carries the genes for most
`or all of its components; any organism lacking the complex
`or pathway often lacks all of the component genes. The co-
`inherited proteins can be identified in an automated
`fashion by comparing their phylogenetic profiles, strings that
`encode the presence or absence of sequence homologs in
`known genomes, as shown for a few proteins in Figure 3.
`
`Each phylogenetic profile is analogous to an abstract rep-
`resentation of an evolutionary tree; matching phylogenetic
`profiles therefore identifies proteins with similar patterns
`of inheritance. Note that no homology is required among
`the proteins with similar phylogenetic profiles; the pro-
`teins are co-inherited and, when many genomes (n > 10)
`are analyzed, usually functionally related, as in the exam-
`ples shown in Figure 3. The involvement of the
`uncharacterized SmpB protein family in protein synthesis,
`
`predicted by phylogenetic profiles [14•], was recently con-
`firmed by Karzai et al. [15].
`
`The differential genome analysis method of Huynen
`et al. [16] also takes advantage of gene presence and
`absence to associate phenotypes with genes: a list is pre-
`pared of genes shared among organisms that also share a
`given phenotype. This list of genes is filtered by removing
`the genes that occur in organisms lacking the phenotype.
`The remaining genes are correspondingly enriched for
`those that confer the phenotype.
`
`Homology (and evolution)
`The distinction between homology and nonhomology
`methods can be blurred, as even direct sequence compar-
`isons are enhanced by taking advantage of evolutionary
`variations. For example, Lichtarg et al. [17] showed that
`functional sites on proteins could be identified by analyz-
`ing amino acids conserved at different branching depths in
`phylogenetic trees of protein homologs. Likewise, varia-
`tions among protein homologs found by clustering the
`proteins in phylogenetic trees often reveal subtle special-
`ization in protein function. Recent examples of this
`
`cele
`tpal
`tmar
`syne
`rpxx
`pyro
`paer
`paby
`mtub
`mthe
`mpne
`mjan
`mgen
`hpy9
`hpyl
`hinf
`ecol
`ctra
`cpne
`bsub
`bbur
`aqua
`aful
`aero
`
`MSH1
`~PMS1
`MLH1
`PMS1
`MSH5
`MSH3
`MSH6
`MSH2
`MLH1
`YPL207W
`RPS4B
`RS4E
`EIF2A
`RNAP27
`R37A
`RPL43B
`RPS6A
`RS6
`RS33
`RS33
`RL31A
`R27A
`ADE13
`PUR2
`PURA
`
`DNA repair
`
`Ribosomal
`
`metabolism
`
`Purine
`
`Figure 3
`
`Phylogenetic profiles [14•] for three groups
`of yeast proteins (ribosomal proteins and
`proteins involved in DNA repair and purine
`metabolism) sharing similar co-inheritance
`patterns. Each row is a graphical
`representation of a protein phylogenetic
`profile, with elements colored according to
`whether a homolog is absent (white box) or
`present (colored box) in each of
`24 genomes (columns). When homology is
`present, the elements are shaded on a
`gradient from light gray (low homology) to
`black (strong homology). In this case,
`homologs are considered absent when no
`BLAST hits [44] are found with expectation
`(E) values < 1 × 10–5. When homologs are
`present, the profile receives a score
`(–1/log E) that describes the degree of
`sequence similarity with the best match in
`that genome. Note that an uncharacterized
`protein (YPL207W) clusters with the
`ribosomal proteins and can now be
`assigned a function in protein synthesis.
`
`Current Opinion in Structural Biology
`
`Petitioner Microsoft Corporation - Ex. 1037, p. 361
`
`

`

`362 Sequences and topology
`
`analysis include the MutS protein family [18] and proteins
`conserved between worm and yeast [19].
`
`A different aspect of evolutionary information is used in
`the calculation of COGS (clusters of orthologs) [20], in
`which proteins from different organisms are grouped
`together in such a way as to maximize their functional
`equivalence. COGS are generated by
`identifying
`orthologs or equivalent proteins among different organ-
`isms. Orthologs can be defined operationally as the
`symmetric top-scoring protein sequences in a sequence
`homology search. That is, a query sequence from
`genome 1 has an ortholog in genome 2 if searching the
`query versus genome 2 turns up the ortholog as the best
`match and searching the ortholog versus genome 1 turns
`up the query protein as the best match.
`
`COGS effectively cluster functionally equivalent pro-
`teins because of the power of orthology, which dictates
`that not only are sequences homologous, but also that
`they are the best homologs regardless of search direction.
`This symmetric homology detection relies on the
`absence of better homologs from each genome and,
`therefore, incorporates both evolutionary information and
`sequence matching. Phylogenetic profiles can be con-
`structed from orthologs, rather than from best homologs,
`and searched for exact matches at the COGS web site
`(http://www.ncbi.nlm.nih.gov/COG/) [20].
`
`No homology required (made up for with extra
`data)
`Each of the methods discussed above requires that a query
`protein have some sequence homologs in the database,
`even though direct sequence homology with these pro-
`teins may not be the basis for the analysis. This
`requirement is lifted for analyses of other genomic data,
`however, such as analysis of correlated mRNA expression lev-
`els, reviewed in [21,22]. Therefore, these techniques can
`find relationships among proteins that are absolutely
`unique. The premise of all expression clustering methods
`is, as in phylogenetic profiles, that proteins rarely work
`alone, but are often expressed at the same time or place as
`functionally related proteins. By varying the conditions
`that cells are grown in or by choosing different cell types or
`cells from different tissues, enough variation in gene
`expression can be observed to identify coexpressing genes.
`
`Such clustering requires additional data beyond genome
`sequences, to date relying on measurements of cellular
`mRNA levels by DNA microarrays, as in [23], serial analy-
`sis of gene expression (SAGE) libraries [24] or expressed
`sequence tag (EST) libraries [25]. Underlying the cluster-
`ing of genes by their mRNA coexpression levels is the
`assumption that coexpressed genes will generally be func-
`tionally related if enough different conditions have been
`tested. Such clustering performs well for strongly coex-
`pressed genes, such as ribosomal subunits, and poorly for
`other gene groups. It requires fairly large sets of data, such
`
`as more than 70 DNA chip measurements of yeast mRNA
`levels [26••,27••] or hundreds of human EST libraries from
`different tissues and cells [28•].
`
`In a manner analogous to analyzing gene co-inheritance
`or mRNA expression patterns, an organism’s proteins can
`probably be clustered effectively by their own protein
`coexpression patterns under varying growth conditions.
`For protein coexpression analysis, one directly measures
`the functional species (proteins) and it is likely that the
`clusters calculated on this basis will be more powerful for
`protein function assignment than mRNA expression
`clustering, especially given that protein and mRNA lev-
`els are often surprisingly uncorrelated [29•]. Protein
`expression patterns have been measured directly by
`mass spectrometry of protein mixtures [30] and by vari-
`ous two-dimensional gel electrophoresis techniques,
`with the proteins on the gel identified by amino acid
`content [31] or mass spectrometry [32•,33•]. The tech-
`niques are labor intensive and have not, just yet,
`produced a sufficient dataset for coexpression analyses.
`This is likely to change in the very near future, given the
`current emphasis on genome-wide and proteome-wide
`analyses. Protein expression patterns have also been
`measured, not as a function of growth conditions, but
`spatially, as β-galactosidase
`fusions
`in Xenopus
`embryos [34•], allowing functionally related proteins to
`be grouped by their spatial coexpression patterns.
`
`Building a genome-wide network of proteins
`The methods described above are easily applied on a
`genome-wide scale, combining results from each method to
`build a network of the functional relationships among an
`organism’s proteins. Such a network was calculated recently
`for yeast proteins [27••], identifying 93,750 functional links
`among 4701 of the 6217 proteins in yeast. A subset of this
`network is drawn in Figure 4, showing the amazing com-
`plexity of the connections generated by these methods.
`Perhaps even more surprising is the high degree of connec-
`tivity among the proteins, attributable in part to homology
`and false-positive predictions, but still observable even in
`entirely experimentally derived networks, such as the con-
`nected set of 542 proteins linked according to 727
`experimentally observed protein–protein interactions from
`the Database of Interacting Proteins (inset, Figure 4) [3•,35].
`These studies reinforce the idea that proteins rarely work in
`isolation, but are instead linked into an interconnected net-
`work of physical interactions and functional relationships.
`
`Why is this computational genetics?
`Unlike sequence homology and inferences from protein
`structure, nonhomology methods reveal protein function in
`the same manner that experimental geneticists do: by defin-
`ing the context that the protein operates in. Function is then
`determined from the pathway neighbors of a protein. For
`this reason, we might consider nonhomology methods to be
`computational genetics, a bioinformatics analysis that proceeds
`in a fashion analogous to experimental genetics.
`
`Petitioner Microsoft Corporation - Ex. 1037, p. 362
`
`

`

`Figure 4
`
`Current Opinion in Structural Biology
`
`The network of 12,012 functional relationships among 2240 proteins
`from yeast generated from protein phylogenetic profiles, showing links
`that occur with an expectation (E) value < 1 × 10–3. Each vertex
`represents a protein and each line represents a functional link,
`modeled as springs to position functionally related proteins close
`together in space [27••]. In this case, the phylogenetic profiles are
`calculated for n = 24 genomes and the E value of a link from protein A
`to protein B is calculated as p(A)•V(n,dAB)•N•C, where p(A) is the
`probability of observing the profile of protein A, V(n,dAB) is the volume
`of an n-dimensional hypersphere centered on A of radius dAB, N is the
`number of proteins with informative vectors and C is a scale factor. For
`comparison, the inset shows an experimentally derived network of
`protein–protein interactions from the Database of Interacting
`Proteins [35], courtesy of I Xenarios.
`
`In fact, the method of phylogenetic profiles [14•] is an
`exact computational equivalent of the experimental
`genetics approach of mapping a mutant gene’s phenotype
`to the gene. When we compare one organism with anoth-
`er, we can generalize each organism as having a collection
`of mutations, gene knockouts and extra genes relative to
`the other organisms. By grouping genes with similar phy-
`logenetic profiles, we are mapping genes that produce
`shared phenotypes (the genes are expressed or absent in
`the same sets of organisms) and are essentially perform-
`ing a standard genetic mapping. Of course, the
`experiment is performed computationally and in a mas-
`sively parallel form, but it is essentially the same analysis
`as in experimental genetics.
`
`Conclusions
`This past year has seen an explosion of new experimental
`and computational tools to identify protein function,
`including the development of ‘nonhomology’ computa-
`tional methods. These methods take advantage of the
`many properties shared among functionally related pro-
`teins, such as patterns of domain fusion, evolutionary
`co-inheritance, conservation of relative gene position and
`
`Nonhomology methods Marcotte 363
`
`correlated expression patterns. Such analyses, building on
`existing genomic sequence and expression data, allow the
`assignment of preliminary protein function on a genome-
`wide scale. Even more exciting is the potential for
`increasing the power of the methods as more genome
`sequences and expression libraries accumulate; for exam-
`ple, the number of possible phylogenetic profile vectors
`and, therefore, the potential to separate unrelated proteins
`grows on the order of 2n for n genomes. An important goal
`will be working out proper statistical evaluations of results
`from each of the methods.
`
`In the next year, we can expect these techniques to be
`integrated, for each genome, with homology- and struc-
`ture-derived protein functions, as well as with known
`experimental data, as researchers are beginning to extract
`experiments from the scientific literature into computer-
`analyzable databases, such as for protein–protein
`interactions [35,36], functional relationships derived from
`the co-occurrence of gene names in articles [37], metabol-
`ic pathways [38,39] and general gene function [40–42].
`Beyond even these, we can expect many new types of
`data, such as the recent genome-wide gene disruption phe-
`notypic studies of yeast [43••] and protein expression
`datasets, that should really open up the power of these
`methods and allow researchers to finely map many of the
`functions and relationships among the genes so tantaliz-
`ingly revealed in each newly sequenced genome.
`
`Acknowledgements
`This work was supported by a Department of Energy/Oak Ridge Institute
`for Science and Education Hollaender Distinguished Postdoctoral
`Fellowship and grants from the DOE. The author would like to thank
`David Eisenberg, Matteo Pellegrini, Michael Thompson, Todd Yeates and
`Ioannis Xenarios for support and fruitful scientific collaboration.
`
`References and recommended reading
`Papers of particular interest, published within the annual period of review,
`have been highlighted as:
`• of special interest
`•• of outstanding interest
`Perou CM, Jeffrey SS, van de Rijn M, Rees CA, Eisen MB, Ross DT,
`Pergamenschikov A, Williams CF, Zhu SX, Lee JC et al.: Distinctive
`gene expression patterns in human mammary epithelial cells
`and breast cancers. Proc Natl Acad Sci USA 1999,
`96:9212-9217.
`
`1.
`
`2. Martzen MR, McCraith SM, Spinelli SL, Torres FM, Field S,
`•
`Grayhack EJ, Phizicky EM: A biochemical genomics approach for
`identifying genes by the activity of their products. Science 1999,
`286:1153-1155.
`The genome-wide expression and purification of yeast proteins for associat-
`ing biochemical activities with specific yeast proteins is described.
`
`4.
`•
`
`3. Marcotte EM, Pellegrini M, Ng H-L, Rice DW, Yeates TO, Eisenberg D:
`•
`Detecting protein function and protein–protein interactions from
`genome sequences. Science 1999, 285:751-753.
`This paper shows that domain fusions can be used to predict functionally
`related and physically interacting proteins.
`Enright AJ, Iliopoulos I, Kyrpides NC, Ouzounis CA: Protein
`interaction maps for complete genomes based on gene fusion
`events. Nature 1999, 402:86-90.
`The authors developed a scoring scheme for domain fusion analysis that
`accurately predicts interacting proteins.
`Park J, Teichmann SA, Hubbard T, Chothia C: Intermediate
`sequences increase the detection of homology between
`sequences. J Mol Biol 1997, 273:249-254.
`
`5.
`
`Petitioner Microsoft Corporation - Ex. 1037, p. 363
`
`

`

`364 Sequences and topology
`
`8.
`
`6. Shirasu K, Lahaye T, Tan M-W, Zhou F, Azevedo C, Schulze-Lefert P:
`A novel class of eukaryotic zinc-binding proteins is required for
`disease resistance signaling in barley and development in
`C. elegans. Cell 1999, 99:355-366.
`7. Dandekar T, Snel B, Huynen M, Bork P: Conservation of gene order:
`•
`a fingerprint of proteins that physically interact. Trends Biochem
`Sci 1998, 23:324-328.
`The authors demonstrated that genes that are found as neighbors in multi-
`ple organisms often encode interacting proteins and suggested using con-
`served gene order to predict protein function.
`Tamames J, Casari G, Ouzounis C, Valencia A: Conserved clusters
`of functionally related genes in two bacterial genomes. J Mol Evol
`1997, 44:66-73.
`9. Overbeek R, Fonstein M, D’Souza M, Pusch GD, Maltsev N: The use
`•
`of gene clusters to infer functional coupling. Proc Natl Acad Sci
`USA 1999, 96:2896-2901.
`The authors describe a method that systematizes the analysis of the conser-
`vation of relative gene position among organisms for finding functionally
`related genes.
`10. Jacob F, Monod J: Genetic regulatory mechanisms in the synthe-
`sis of proteins. J Mol Biol 1961, 3:318-356.
`11. Laird AJ, Young IG: Tn5 mutagenesis of the enterochelin gene
`cluster of Escherichia coli. Gene 1980, 11:359-366.
`
`12. Hallas C, Pekarsky Y, Itoyama T, Varnum J, Bichi R, Rothstein JL,
`Croce CM: Genomic analysis of human and mouse TCL1 loci
`reveals a complex of tightly clustered genes. Proc Natl Acad Sci
`USA 1999, 96:14418-14423.
`13. Wu Q, Maniatis T: A striking organization of a large family of human
`neural cadherin-like cell adhesion genes. Cell 1999, 97:779-790.
`
`14. Pellegrini M, Marcotte EM, Thompson MJ, Eisenberg D, Yeates TO:
`•
`Assigning protein functions by comparative genome analysis: protein
`phylogenetic profiles. Proc Natl Acad Sci USA 1999, 96:4285-4288.
`This paper shows that functional relationships among proteins can be iden-
`tified by analyzing protein co-inheritance.
`15. Karzai AW, Susskind MM, Sauer RT: SmpB, a unique RNA-binding
`protein essential for the peptide-tagging activity of SsrA (tmRNA).
`EMBO J 1999, 18:3793-3799.
`16. Huynen M, Dandekar T, Bork P: Differential genome analysis
`applied to the species-specific features of Helicobacter pylori.
`FEBS Lett 1998, 426:1-5.
`17. Lichtarg O, Bourne HR, Cohen FE: An evolutionary trace method
`defines binding surfaces common to protein families. J Mol Biol
`1996, 257:342-358.
`18. Eisen JA: A phylogenomic study of the MutS family of proteins.
`Nucleic Acids Res 1998, 26:4291-4300.
`
`19. Chervitz SA, Aravind L, Sherlock G, Ball CA, Koonin EV, Dwight SS,
`Harris MA, Dolinski K, Mohr S, Smith T et al.: Comparison of the
`complete protein sets of worm and yeast: orthology and
`divergence. Science 1998, 282:2022-2028.
`20. Tatusov RL, Koonin EV, Lipman DJ: A genomic perspective on
`protein families. Science 1997, 278:631-637.
`21. Zhang MQ: Large-scale gene expression data analysis: a new
`challenge to computational biologists. Genome Res 1999, 9:681-688.
`22. Brown PO, Botstein D: Exploring the new world of the genome
`with DNA microarrays. Nat Genet 1999, 21:33-37.
`
`23. Lashkari DA, De Risi JL, McCusker JH, Namath AF, Gentile C,
`Hwang SY, Brown PO, Davis RW: Yeast microarrays for genome
`wide parallel genetic and gene expression analysis. Proc Natl
`Acad Sci USA 1997, 94:13057-10362.
`24. Velculescu VE, Zhang L, Vogelstein B, Kinzler KW: Serial analysis of
`gene expression. Science 1995, 270:484-487.
`
`25. Adams MD, Kelley JM, Gocayne JD, Dubnick M, Polymeropoulos MH,
`Xiao H, Merril CR, Wu A, Olde B, Moreno RF et al.: Complementary
`DNA sequencing: expressed sequence tags and human genome
`project. Science 1991, 252:1651-1656.
`26. Eisen MB, Spellman PT, Brown PO, Botstein D: Cluster analysis and
`••
`display of genome-wide expression patterns. Proc Natl Acad Sci
`USA 1998, 95:14863-14868.
`The authors recognized that functionally related genes could be grouped by
`their mRNA expression patterns and introduced a visual presentation that
`has now become common in the field.
`
`27. Marcotte EM, Pellegrini M, Thompson MJ, Yeates TO, Eisenberg D:
`••
`A combined algorithm for genome-wide prediction of protein
`function. Nature 1999, 402:83-86.
`A synthesis of nonhomology methods allowing genome-wide prediction of
`protein function in yeast.
`28. Walker MG, Volkmuth W, Klingler TM: Pharmaceutical target
`•
`discovery using Guilt-by-Association: schizophrenia and
`Parkinson’s disease genes. In Proceedings of the 7th International
`Conference on Intelligent Systems for Molecular Biology: 1999
`August 6–10; Heidelberg, Germany. Edited by Lengauer T,
`Schneider R, Bork P, Brutlag D, Glasgow J, Mewes H-W, Zimmer R.
`Menlo Park, California: AAAI Press; 1999:282-286.
`A method for discovering functionally related genes by analysis of their co-
`occurrence in expressed sequence tag libraries is described.
`29. Gygi SP, Rochon Y, Franza BR, Aebersold R: Correlation between
`•
`protein and mRNA abundance in yeast. Mol Cell Biol 1999,
`19:1720-1730.
`This paper provides the interesting observation that protein levels and mRNA
`levels are often poorly correlated, often varying by up to 30-fold in a test of
`about 300 sequences, providing the promise that protein coexpression may
`surpass mRNA coexpression for clustering genes of related function.
`30. Ducret A, Oostveen IV, Eng JK, Yates JR, Aebersold R: High
`throughput protein characterization by automated reverse-phase
`chromatography/electrospray tandem mass spectrometry. Protein
`Sci 1998, 7:706-719.
`
`31. Garrels JI, Futcher B, Kobayashi R, Latter GI, Schwender B, Volpe T,
`Warner JR, McLaughlin CS: Protein identification for a
`Saccharomyces cerevisiae protein database. Electrophoresis 1994,
`15:1466-1486.
`
`32. Neubauer G, King A, Rappsilber J, Calvio C, Watson M, Ajuh P,
`•
`Sleeman J, Lamond A, Mann M: Mass spectrometry and EST-
`database searching allows characterization of the multi-protein
`spliceosome complex. Nat Genet 1998, 20:46-50.
`A model example of the rapid identification of components of protein
`complexes.
`
`33. Gygi SP, Rist B, Gerber SA, Turacek F, Gelb MH, Aebersold R:
`•
`Quantitative analysis of complex protein mixtures using isotope-
`coded affinity tags. Nat Biotech 1999, 17:994-999.
`A potentially important paper describing the accurate, high-throughput mea-
`surement of protein expression patterns.
`
`34. Gawantka V, Pollet N, Delius H, Vingron M, Pfister R, Nitsch R,
`•
`Blumenstock C, Niehrs C: Gene expression screening in Xenopus
`identifies molecular pathways, predicts gene function and
`provides a global view of embryonic patterning. Mech Dev 1998,
`77:95-141.
`This paper demonstrates the analysis of the tissue expression patterns of
`1765 cDNAs by in situ hybridization in Xenopus embryos; each gene’s
`expression pattern is photographed and cataloged in a database, allowing
`the clustering of genes with similar spatial expression patterns.
`
`35. Xenarios I, Rice DW, Salwinski L, Baron MK, Marcotte EM,
`Eisenberg D: DIP: the Database of Interacting Proteins. Nucleic
`Acids Res 2000, 28:289-291.
`36. Blaschke C, Andrade MA, Ouzounis C, Valencia A: Automatic
`extraction of biological information from scientific text:
`protein–protein interactions. In Proceedings of the 7th International
`Conference on Intelligent Systems for Molecular