I-BBC yase-United Kingdom namuhla (mhla ziyi-7 kuNcwaba) ishicilele umbiko othi abacwaningi baseStanford University e-United States basebenzise ubuchwepheshe be-AI ukuze bakhe elabhorethri igciwane elisha elingaphindaphindeka futhi elisebenza ngokuphelele.
Abacwaningi bathi lokhu kungokokuqala i-AI ikwazi ngempumelelo ukuklama i-genome ephelele. Izinhlobo ezintsha zamagciwane ezingu-16 ezikhiqiziwe zisetshenziselwa ukuthelela amagciwane, futhi azibeki usongo kubantu.
Ngesikhathi exoxisana ne-BBC, uBrian Hie, uprofesa osizayo eStanford University, wathi i-AI yayisike yasetshenziselwa ukuklama ama-antibiotic amasha, kodwa ngokombono wezobuchwepheshe, ukuklama igciwane eliphilayo kusukela ekuqaleni kunzima kakhulu.
Iqembu labacwaningi lathi indlela elisebenza ngayo ifana namamodeli amakhulu olimi afana ne-ChatGPT, kodwa amamodeli amakhulu olimi abikezela ukulandelana kwemibhalo, kanti amamodeli e-Evo1 ne-Evo2 athuthukiswe futhi asetshenziswa yileli qembu abikezela “ulimi lokuphila”.
Mayelana nokuqeqeshwa kwamamodeli, iqembu labacwaningi lathi idatha efanele ivela kumakhodi ofuzo amagciwane, amabhaktheriya, izitshalo nabantu. Ngemuva kwalokho iqembu lasebenzisa ukulungisa kahle (Fine-tuning) amamodeli ukuze aklame ama-bacteriophage (amagciwane athelela kuphela izinhlobo ezithile zamabhaktheriya).
Qaphela: Kumamodeli amakhulu olimi, lokhu kusho ukusebenzisa iqoqo ledatha eliqondene nomsebenzi ukuze kuqeqeshwe ngokwengeziwe imodeli enkulu yolimi esivele yaqeqeshwa. Imodeli eqeqeshwe kusengaphambili isuke isifunde ulwazi oluningi olujwayelekile, kanti ukulungisa kahle kuyisiza ukuthi igxile emkhakheni othile.
Ukuqinisekisa ukuthi imiklamo ye-AI iyasebenza yini, abacwaningi baseStanford bakhetha imiklamo ye-AI engu-302 eyayithembisa kakhulu bayikhiqiza elabhorethri. Imiphumela yabonisa ukuthi imiklamo engu-16 kuyo yayikwazi ukubulala i-Escherichia coli (E. coli) ngempumelelo.
Ukuthuthukiswa kwama-bacteriophage amasha kunganikeza indlela entsha yokwelapha izifo ezibangelwa amagciwane angasazweli kuma-antibiotic. Lo mbiko uveza ukuthi ukwelashwa ngama-phage (phage therapy) kubhekwa njengendlela engase isetshenziswe ukubhekana nokwanda kwezifo ezibangelwa amabhaktheriya, ikakhulukazi izifo ezingelapheki ngama-antibiotic.
Le mpumelelo iphinde ibonise ukuthi i-AI isinamandla okuklama amasistimu amasha okuphila angatholakali emvelweni. UHie ukholelwa ukuthi leli khono lingayithuthukisa kakhulu impilo yabantu ngokusungula imithi emisha nezindlela zokwelapha ezintsha.
Amakhodi ofuzo ama-bacteriophage kulolu cwaningo anobude obungaba ngu-5,400 base pairs. Uma kuqhathaniswa, i-genome yeseli eliphilayo elincane kunazo zonke inama-base pairs angaba ngu-500,000, kanti i-genome yomuntu inama-base pairs angu-3,000,000,000.
UHie wathi ukuzama ukuklama ezinye izinto eziphilayo ezilula “kungadinga umsebenzi omningi, kodwa akuyona into engenakwenzeka”. Ubuye wathi iqembu “linentshisekelo yangempela yokuqhubekela kule ndlela”.
UMarc Güell, uprofesa waselabhorethri ye-synthetic biology ePompeu Fabra University eSpain, wathi lolu cwaningo “luyisigameko esibaluleke kakhulu”, ngoba “ngokokuqala emlandweni, abantu sebeqala ukuklama ibhayoloji kumakhompyutha”.
Uthe lokhu kwenza abantu bakwazi ukucabanga ngamathuba amaningi okubhekana nezinselele ezinkulu zesintu, okuhlanganisa ama-bacteriophage okwelapha izifo, ama-enzyme okwelapha izifo zofuzo, kanye nama-antibody angasetshenziswa ekwelashweni okusebenzisa amasosha omzimba.
Inkomba
Generative design of bacteriophages with genome language models