Initial sequencing and analysis of the human genome - Vitagenes

articlesoverprediction rate of 30% for gene predictions in this expanded set,the analysis above suggests that IGI+ set contains about 28,000 truegenes and yields an estimate of about 32,000 human genes. We areinvestigating ways to ®lter the expanded set, to produce an IGI withthe advantage of the increased sensitivity resulting from combiningmultiple gene prediction programs without the corresponding lossof speci®city. Meanwhile, the IGI+ set can be used by researcherssearching for genes that cannot be found in the IGI.Some classes of genes may have been missed by all of the gene-®nding methods. Genes could be missed if they are expressed at lowlevels or in rare tissues (being absent or very under-represented inEST and mRNA databases) and have sequences that evolve rapidly(being hard to detect by protein homology and genome comparison).Both the worm and ¯y gene sets contain a substantial numberof such genes 293,294 . Single-exon genes encoding small proteins mayalso have been missed, because EST evidence that supports themcannot be distinguished from genomic contamination in the ESTdataset and because homology may be hard to detect for smallproteins 310 .The human thus appears to have only about twice as many genesas worm or ¯y. However, human genes differ in important respectsfrom those in worm and ¯y. They are spread out over much largerregions of genomic DNA, and they are used to construct morealternative transcripts. This may result in perhaps ®ve times as manyprimary protein products in the human as in the worm or ¯y.The predicted gene and protein sets described here are clearly farfrom ®nal. Nonetheless, they provide a valuable starting point forexperimental and computational research. The predictions willimprove progressively as the sequence is ®nished, as furthercon®rmatory evidence becomes available (particularly fromother vertebrate genome sequences, such as those of mouse andT. nigroviridis), and as computational methods improve. We intendto create and release updated versions of the IGI and IPI regularly,until they converge to a ®nal accurate list of every human gene. Thegene predictions will be linked to RefSeq, HUGO and SWISSPROTidenti®ers where available, and tracking identi®ers between versionswill be included, so that individual genes under study can be tracedforwards as the human sequence is completed.Comparative proteome analysisKnowledge of the human proteome will provide unprecedentedopportunities for studies of human gene function. Often clues willbe provided by sequence similarity with proteins of known functionin model organisms. Such initial observations must then be followedup by detailed studies to establish the actual function of thesemolecules in humans.For example, 35 proteins are known to be involved in the vacuolarprotein-sorting machinery in yeast. Human genes encoding homologuescan be found in the draft human sequence for 34 of theseyeast proteins, but precise relationships are not always clear. In ninecases there appears to be a single clear human orthologue (a genethat arose as a consequence of speciation); in 12 cases there arematches to a family of human paralogues (genes that arose owing tointra-genome duplication); and in 13 cases there are matchesto speci®c protein domains 311±314 . Hundreds of similar storiesemerge from the draft sequence, but each merits a detailed interpretationin context. To treat these subjects properly, there will bemany following studies, the ®rst of which appear in accompanyingpapers 315±323 .Here, we aim to take a more global perspective on the content ofthe human proteome by comparing it with the proteomes of yeast,worm, ¯y and mustard weed. Such comparisons shed useful light onthe commonalities and differences among these eukaryotes 294,324,325 .The analysis is necessarily preliminary, because of the imperfectnature of the human sequence, uncertainties in the gene and proteinsets for all of the multicellular organisms considered and ourincomplete knowledge of protein structures. Nonetheless, somegeneral patterns emerge. These include insights into fundamentalmechanisms that create functional diversity, including invention ofprotein domains, expansion of protein and domain families, evolutionof new protein architectures and horizontal transfer of genes.Other mechanisms, such as alternative splicing, post-translationalmodi®cation and complex regulatory networks, are also crucial ingenerating diversity but are much harder to discern from theprimary sequence. We will not attempt to consider the effects ofalternative splicing on proteins; we will consider only a single spliceform from each gene in the various organisms, even when multiplesplice forms are known.Functional and evolutionary classi®cation. We began by classifyingthe human proteome on the basis of functional categories andevolutionary conservation. We used the InterPro annotation protocolto identify conserved biochemical and cellular processes.InterPro is a tool for combining sequence-pattern informationfrom four databases. The ®rst two databases (PRINTS 326 andProsite 327 ) primarily contain information about motifs correspondingto speci®c family subtypes, such as type II receptor tyrosinekinases (RTK-II) in particular or tyrosine kinases in general. Thesecond two databases (Pfam 307 and Prosite Pro®le 327 ) containinformation (in the form of pro®les or HMMs) about families ofstructural domainsÐfor example, protein kinase domains. Inter-Pro integrates the motif and domain assignments into a hierarchicalclassi®cation system; so a protein might be classi®ed at the mostdetailed level as being an RTK-II, at a more general level as being akinase speci®c for tyrosine, and at a still more general level asbeing a protein kinase. The complete hierarchy of InterPro entriesis described at http://www.ebi.ac.uk/interpro/. We collapsed theInterPro entries into 12 broad categories, each re¯ecting a set ofcellular functions.The InterPro families are partly the product of human judgementand re¯ect the current state of biological and evolutionary knowledge.The system is a valuable way to gain insight into largecollections of proteins, but not all proteins can be classi®ed atpresent. The proportions of the yeast, worm, ¯y and mustard weedprotein sets that are assigned to at least one InterPro family is, foreach organism, about 50% (Table 23; refs 307, 326, 327).About 40% of the predicted human proteins in the IPI could beassigned to InterPro entries and functional categories. On the basisof these assignments, we could compare organisms according to thenumber of proteins in each category (Fig. 37). Compared with thetwo invertebrates, humans appear to have many proteins involvedin cytoskeleton, defence and immunity, and transcription andtranslation. These expansions are clearly related to aspects ofvertebrate physiology. Humans also have many more proteins thatare classi®ed as falling into more than one functional category (426in human versus 80 in worm and 57 in ¯y, data not shown).Interestingly, 32% of these are transmembrane receptors.We obtained further insight into the evolutionary conservation ofproteins by comparing each sequence to the complete nonredundantdatabase of protein sequences maintained at NCBI, using theBLASTP computer program 328 and then breaking down the matchesaccording to organismal taxonomy (Fig. 38). Overall, 74% of theproteins had signi®cant matches to known proteins.Such classi®cations are based on the presence of clearly detectablehomologues in existing databases. Many of these genes have surelyevolved from genes that were present in common ancestors but havesince diverged substantially. Indeed, one can detect more distantrelationships by using sensitive computer programs that can recognizeweakly conserved features. Using PSI-BLAST, we can recognizeprobable nonvertebrate homologues for about 45% of the `vertebrate-speci®c'set. Nonetheless, the classi®cation is useful for gaininginsights into the commonalities and differences among theproteomes of different organisms.Probable horizontal transfer. An interesting category is a set of 223proteins that have signi®cant similarity to proteins from bacteria,but no comparable similarity to proteins from yeast, worm, ¯y andNATURE | VOL 409 | 15 FEBRUARY 2001 | www.nature.com 901© 2001 Macmillan Magazines Ltd