Eukaryotic promoter recognition.

On ly recen tly h as it becom e common to determ in e eukaryotic gen om ic sequen ces large en ough to con tain several gen es. With th ese data com es a n ew problem for gen e fin din g program s: to partit ion a set of exon s correctly am on g several gen es. On e lin e of developm en t in eukaryot ic gen e iden tification begin s with codin g region iden tificat ion by statist ical m ean s an d adds pattern recogn it ion for sites of tran scrip tion al, sp licin g, an d tran slat ion al con trol to produce algorithm s capable of suggest in g overall gen e structu re (for review, see Gelfan d 1995; Fickett 1996a). To date, m ost developm en t effort h as focused on in tegrat ion of th e various kin ds of pattern in form ation in th e relat ively sim ple case wh ere a sin gle complete gen e is presen t in th e in put sequen ce. In th is case, curren t algorithm s usually suggest a putative protein tran slat ion sim ilar to th at in th e literature, th ough th ere is st ill sign ifican t room for im provem en t (Burset an d Guigo 1996). Th e exten sion of th ese algorithm s to deal with a sequen ce con tain in g multip le or part ial gen es is just begin n in g (Burge an d Karlin 1997; h t t p :/ / gn om ic.st an fo rd .ed u / ∼ch ris/ GENSCANW.h tm l). Because th e sign als th at con trol th e start an d stop of tran scrip tion an d tran slation , an d th e location of splicin g, are still n ot very well un derstood, it is n ot un common for a gen e-fin din g algorithm to con fuse in tern al with in it ial an d term in al exon s, th us wron gly part it ion in g th e exon s. Th e problem is compoun ded by our in complete un derstan din g of altern ative splicin g con trol elem en ts. An oth er lin e of developm en t in gen e iden tificat ion is based on h omology (e.g., Gish an d States 1993; Gelfan d et al. 1996). If th ere is a close h omolog in th e databases to on e of th e gen es in th e sequen ce un der an alysis, sequen ce sim ilarity will usually group th e exon s for th is gen e correctly. Still, in m an y cases th ere is n o close h omolog an d n o guaran tee wh en th ere is som e h om olog th at th e en coded protein lacks in sert ion s/deletion s. Clearly, som e m ean s of recogn izin g th e begin n in gs of gen es, probably via th e prom oter, or th e en ds, probably by m ean s of th e polyaden ylation sign al or tran slation term in ation sign al (e.g., Kon drakh in et al. 1994; Wah le an d Keller 1996; Dalph in et al. 1997; Solovyev an d Salam ov 1997), would en able a m ajor advan ce. Th e prom oter seem s to be a m uch rich er sign al th an th e 38 processin g sign als, th ough , as we sh all see below, it is n ot easy to take advan tage of th e in form ation in th e prom oter.