izindaba

I-Lasker Basic Medical Research Award yalo nyaka iklonyeliswe u-Demis Hassabis no-John Jumper ngeqhaza labo ekwakhiweni kohlelo lobuhlakani bokwenziwa lwe-AlphaFold olubikezela ukwakheka kwezingxenye ezintathu zamaprotheni okusekelwe ekulandeleni kokuqala kwama-amino acid.

Imiphumela yabo ixazulula inkinga osekunesikhathi ikhathaza umphakathi wezesayensi futhi ivula umnyango wokusheshisa ucwaningo kuwo wonke umkhakha wezokwelapha. Amaprotheni adlala indima ebalulekile ekuthuthukisweni kwezifo: esifweni i-Alzheimer's, ayagoqa futhi ahlangane; Ngomdlavuza, umsebenzi wabo wokulawula ulahlekile; Ezinkingeni ze-metabolic ezizalwa nazo, azisebenzi kahle; Ku-cystic fibrosis, bangena endaweni engafanele esitokisini. Lezi ezinye zezindlela ezimbalwa kweziningi ezibangela izifo. Amamodeli anemininingwane esakhiwo samaprotheni angahlinzeka ngokucushwa kwe-athomu, aqhube ukwakheka noma ukukhethwa kwama-molecule ahlobene kakhulu, futhi asheshise ukutholwa kwezidakamizwa.

Izakhiwo zamaprotheni ngokuvamile zinqunywa nge-X-ray crystallography, i-nuclear magnetic resonance kanye ne-cryo-electron microscopy. Lezi zindlela ziyabiza futhi zidla isikhathi. Lokhu kubangela isizindalwazi esikhona sesakhiwo samaprotheni e-3D esinedatha yesakhiwo engaba ngu-200,000 kuphela, kuyilapho ubuchwepheshe bokulandelana kwe-DNA bukhiqize ukulandelana kwamaprotheni angaphezu kwezigidi ezingu-8. Ngawo-1960, u-Anfinsen et al. ithole ukuthi ukulandelana kwe-1D kwama-amino acid kungagoqeka ngokuzenzakalelayo nangokuphindaphindiwe kube ukuhambisana okunezinhlangothi ezintathu ezisebenzayo (Umfanekiso 1A), nokuthi “ama-chaperone” amangqamuzana angakwazi ukusheshisa futhi enze le nqubo ibe lula. Lokhu kubuka kuholela enseleleni yeminyaka engama-60 kubhayoloji yamangqamuzana: ukubikezela ukwakheka kwe-3D kwamaprotheni kusukela ku-1D yokulandelana kwama-amino acid. Ngokuphumelela kwe-Human Genome Project, ikhono lethu lokuthola ukulandelana kwe-amino acid ye-1D libe ngcono kakhulu, futhi le nselele isiphuthuma nakakhulu.

Ukubikezela izakhiwo zamaprotheni kunzima ngezizathu eziningana. Okokuqala, zonke izindawo ezinezinhlangothi ezintathu ezingaba khona ze-athomu ngayinye kuwo wonke ama-amino acid zidinga ukuhlola okuningi. Okwesibili, amaprotheni asebenzisa kakhulu ukuhambisana esakhiweni samakhemikhali awo ukuze amise kahle ama-athomu. Njengoba amaprotheni ngokuvamile enamakhulu “abanikeli” be-hydrogen bond (imvamisa i-oksijini) okufanele ibe seduze nesibopho se-hydrogen “sokwamukela” (imvamisa i-nitrogen eboshelwa ku-hydrogen), kungaba nzima kakhulu ukuthola ama-conformations lapho cishe wonke umnikezeli eseduze nomamukeli. Okwesithathu, kunezibonelo ezilinganiselwe zokuqeqeshwa kwezindlela zokuhlola, ngakho-ke kuyadingeka ukuqonda ukusebenzisana okungaba khona kwezinhlangothi ezintathu phakathi kwama-amino acid ngesisekelo sokulandelana kwe-1D kusetshenziswa ulwazi mayelana nokuvela kwamaprotheni afanelekile.

I-Physics yaqale yasetshenziswa ukumodela ukusebenzisana kwama-athomu ekufuneni ukuhambisana okungcono kakhulu, futhi kwasungulwa indlela yokubikezela ukwakheka kwamaphrotheni. U-Karplus, u-Levitt no-Warshel baklonyeliswe ngomklomelo we-Nobel we-Chemistry ngo-2013 ngomsebenzi wabo wokulingisa amaprotheni. Kodwa-ke, izindlela ezisekelwe ku-physics zibiza ngokwezibalo futhi zidinga ukucubungula cishe, ngakho-ke izakhiwo ezinezinhlangothi ezintathu ezinembayo azikwazi ukubikezelwa. Enye indlela “esekelwe olwazini” iwukusebenzisa imininingwane yolwazi yezakhiwo ezaziwayo nokulandelanisa ukuqeqesha amamodeli ngobuhlakani bokwenziwa nokufunda ngomshini (AI-ML). I-Hassabis ne-Jumper zisebenzisa izici zakho kokubili i-physics ne-AI-ML, kodwa ukuqanjwa nokugxumela ekusebenzeni kwendlela kuvela ngokuyinhloko ku-AI-ML. Laba bacwaningi ababili ngobuchule bahlanganisa isizindalwazi esikhulu somphakathi nezinsiza zekhompuyutha zebanga lezimboni ukuze bakhe i-AlphaFold.

Sazi kanjani ukuthi "baxazulule" indida yokubikezela kwesakhiwo? Ngo-1994, kwasungulwa umncintiswano Wokuhlola Okubalulekile Kokubikezelwa Kwesakhiwo (CASP), ohlangana njalo ngemva kweminyaka emibili ukulandelela inqubekelaphambili yokubikezela kwesakhiwo. Abacwaningi bazokwabelana ngokulandelana kwe-1D kwephrotheni isakhiwo sabo abasanda kusixazulula, kodwa imiphumela yabo engakashicilelwa. Umbikezeli ubikezela ukwakheka kwezinhlangothi ezintathu kusetshenziswa lokhu kulandelana kwe-1D, futhi umhloli wehlulela ngokuzimela ikhwalithi yemiphumela ebikezelwe ngokuyiqhathanisa nesakhiwo esinezinhlangothi ezintathu esinikezwe umhloli (ohlinzekwe kuphela kumhloli). I-CASP yenza ukubuyekezwa okuyimpumputhe kwangempela futhi irekhode ukweqa kokusebenza kwezikhathi ezithile okuhlobene nokuqanjwa kwezindlela. Engqungqutheleni ye-CASP ye-14 ngo-2020, imiphumela yokubikezela kwe-AlphaFold yabonisa ukweqa okukhulu ekusebenzeni kangangokuthi abahleli bamemezela ukuthi inkinga yokubikezela isakhiwo se-3D isixazululiwe: ukunemba kwezibikezelo eziningi kwakusondele kulokho kwezilinganiso zokuhlola.

Ukubaluleka okubanzi ukuthi umsebenzi kaHassabis noJumper ukhombisa ngokugculisayo ukuthi i-AI-ML ingayiguqula kanjani isayensi. Ucwaningo lwayo lubonisa ukuthi i-AI-ML ingakha imibono eyinkimbinkimbi yesayensi evela emithonjeni eminingi yedatha, ukuthi izindlela zokunaka (ezifana nalezo eziku-ChatGPT) zingathola ukuncika okubalulekile nokuhlotshaniswa kwemithombo yedatha, nokuthi i-AI-ML ingakwazi ukuzinqumela yona ngokwayo ikhwalithi yemiphumela yayo yokuphuma. I-AI-ML empeleni yenza isayensi.