iindaba

Lo nyaka iLasker Basic Medical Research Award yawongwa uDemis Hassabis noJohn Jumper ngenxa yegalelo labo ekudaleni inkqubo yengqondo yokwenziwa yeAlphaFold eqikelela ukwakheka kweeproteni ezimacala amathathu esekelwe kulandelelwano lokuqala lweeamino acid.

Iziphumo zabo zisombulula ingxaki ekudala ikhathaza uluntu lwezenzululwazi kwaye ivula ucango lokukhawulezisa uphando kulo lonke icandelo le-biomedical. Iiprotheyini zidlala indima ebalulekileyo ekuphuhlisweni kwezifo: kwisifo sika-Alzheimer, ziyagoqa kwaye zidibanise; Kumhlaza, umsebenzi wabo wokulawula ulahlekile; Kwiingxaki zemetabolism ezizalwa, azisebenzi; Kwi-cystic fibrosis, bangena kwindawo engafanelekanga kwiseli. Ezi zezinye nje zeendlela ezininzi ezibangela izifo. Iimodeli zeprotheyini ezineenkcukacha ezinokubonelela ngolungelelwaniso lwe-athomu, ukuqhuba uyilo okanye ukukhethwa kweeathomu eziphezulu, kunye nokukhawulezisa ukufunyanwa kweziyobisi.

Ulwakhiwo lweprotheyini ngokubanzi lumiselwa yiX-reyi crystallography, i-nuclear magnetic resonance kunye ne-cryo-electron microscopy. Ezi ndlela ziyabiza kwaye zitya ixesha. Oku kubangela i-database ye-3D yesakhiwo seprotheyini ekhoyo kuphela malunga ne-200,000 yedatha yesakhiwo, ngelixa iteknoloji yokulandelelana kwe-DNA ivelise ngaphezu kwe-8 yezigidi zeeprotheyini ezilandelelanayo. Kwiminyaka yoo-1960, uAnfinsen et al. ufumanise ukuba ulandelelwano lwe-1D lwe-amino acids luyakwazi ukukhawuleza kwaye luphinde luphinde lube yinto esebenzayo ye-three-dimensional conformation (Figure 1A), kwaye i-molecular "chaperones" inokukhawulezisa kwaye iququzelele le nkqubo. Olu qwalaselo lukhokelela kumngeni weminyaka engama-60 kwibhayoloji ye-molekyuli: ukuqikelela ukwakheka kwe-3D yeeprotheni ukusuka kwi-1D yokulandelelana kwe-amino acids. Ngempumelelo yeProjekthi yeGenome yoMntu, ukukwazi kwethu ukufumana i-1D amino acid ulandelelwano luphucuke kakhulu, kwaye lo mngeni uye wangxamiseka ngakumbi.

Ukuqikelela izakhiwo zeprotheni kunzima ngenxa yezizathu ezininzi. Okokuqala, zonke iindawo ezinokwakheka-ntathu ezinokubakho kuzo zonke iathom kuzo zonke i-amino acid zifuna ukuphononongwa okuninzi. Okwesibini, iiprotheyini zisebenzisa kakhulu ukuhambelana kwisakhiwo sazo sekhemikhali ukucwangcisa ngokufanelekileyo iiathomu. Ekubeni iiprotheyini zihlala zinamakhulu e-hydrogen bond "abanikeli" (ngokuqhelekileyo i-oksijini) ekufuneka isondele kwi-hydrogen bond "i-acceptor" (ngokuqhelekileyo i-nitrogen ibotshelelwe kwi-hydrogen), kunokuba nzima kakhulu ukufumana ukuhambelana apho phantse wonke umnikezeli esondele kumamkeli. Okwesithathu, kukho imizekelo elinganiselweyo yoqeqesho lweendlela zokuvavanya, ngoko ke kuyimfuneko ukuqonda intsebenziswano enokuthi i-dimensional emithathu phakathi kwe-amino acids ngokwesiseko sokulandelelana kwe-1D usebenzisa ulwazi malunga nokuvela kweeprotheni ezifanelekileyo.

I-Physics yaqala ukusetyenziswa ukulinganisa intsebenziswano yee-athom ekukhangeleni eyona nto ihambelanayo, kwaye kwaphuhliswa indlela yokuqikelela ubume beeprotheni. UKarplus, uLevitt kunye noWarshel bawongwa ngeBhaso leNobel kwiChemistry ngo-2013 ngomsebenzi wabo wokulinganisa iiproteni. Nangona kunjalo, iindlela ezisekwe kwifiziksi zibiza kakhulu kwaye zifuna uqikelelo lokusetyenzwa, ngoko ke ulwakhiwo oluchanekileyo olunemigangatho emithathu alunakuqikelelwa. Enye indlela "esekelwe kulwazi" kukusebenzisa i-database yezakhiwo ezaziwayo kunye nolandelelwano lokuqeqesha iimodeli ngokusebenzisa ubukrelekrele bokwenziwa kunye nokufunda ngomatshini (AI-ML). I-Hassabis kunye ne-Jumper zisebenzisa izinto ze-physics kunye ne-AI-ML, kodwa ukutsha kunye nokutsiba ekusebenzeni kwendlela kubangelwa ikakhulu kwi-AI-ML. Abaphandi babini badibanisa ngokuyilayo i-database enkulu yoluntu kunye nezixhobo zekhompyutha ze-industrial-grade ukwenza i-AlphaFold.

Sazi njani ukuba "bayisombulule" iphazili yengqikelelo yesakhiwo? Ngo-1994, ukhuphiswano lwe-Critical Assessment of Structure Prediction (CASP) lwasekwa, oludibana rhoqo emva kweminyaka emibini ukulandelela inkqubela-phambili yoqikelelo lwesakhiwo. Abaphandi baya kubelana ngokulandelelana kwe-1D yeprotheni enesakhiwo abasandul 'ukusombulula, kodwa iziphumo zabo azikapapashwa. I-predictor iqikelela i-trime-dimensional structure isebenzisa olu luhlu lwe-1D, kwaye umvandlakanyi ugweba ngokuzimeleyo umgangatho weziphumo eziqikelelweyo ngokuzithelekisa nesakhiwo esine-dimensional ezintathu esinikezwe ngumvavanyi (onikezelwe kuphela kumvavanyi). I-CASP iqhuba uphononongo oluyinyani oluyimfama kwaye irekhode imitsi yentsebenzo yamaxesha athile enxulunyaniswa nokuveliswa kwezinto ezintsha. KwiNkomfa ye-14 ye-CASP ngo-2020, iziphumo zengqikelelo ye-AlphaFold zabonisa umtsi kwintsebenzo kangangokuba abaququzeleli babhengeza ukuba ingxaki yoqikelelo lwesakhiwo se-3D isonjululwe: ukuchaneka koqikelelo oluninzi kwakukufutshane naleyo yemilinganiselo yovavanyo.

Ukubaluleka okubanzi kukuba umsebenzi kaHassabis kunye noJumper ubonisa ngendlela ekholisayo indlela i-AI-ML enokuguqula ngayo isayensi. Uphando lwayo lubonisa ukuba i-AI-ML inokwakha iingqikelelo ezintsonkothileyo zesayensi ezivela kwimithombo yedatha emininzi, ukuba iindlela zokungqwalasela (ezifana nezo ze-ChatGPT) zinokufumana ukuxhomekeka okuphambili kunye nokulungelelaniswa kwimithombo yedatha, kwaye i-AI-ML inokuzigweba ngokwayo umgangatho weziphumo zayo eziphumayo. I-AI-ML yenza isayensi ngokwenyani.