Isifinyezo: Abacwaningi bathuthukise futhi baqeqesha inoveli imodeli yokufunda ejulile ebizwa ngokuthi I-BINND (I-Binding and Interaction Neural Network for DNA). Inikwe amandla isethi yedatha engakaze ibonwe yamapheya alandelanayo ayizigidi eziyi-144, i-BINND ibikezela ukuhlangana okubophayo kwe-DNA-DNA ngokunemba okuvelele, okudlula amamodeli angaphambili asezingeni eliphezulu nokusungula ithuluzi elibalulekile lokukhuphula ikhompuyutha esekelwe ku-DNA kanye nokutholwa kwedatha.
Amaqiniso Abalulekile
- I-Hyperconnectivity Barrier Ephukile: Amamodeli endabuko aphatha ukubophezela kwe-DNA njengokuxhumana okulula, okuhlukile kokuthi “Yebo/Cha”. I-BINND yakhelwe ngokukhethekile ukuphatha amanethiwekhi “axhumeke kakhulu”—ibikezela ukuthi imicu eminingi ye-DNA ixhumana kanjani ngesikhathi esisodwa, ilingisa indawo eminyene yengqamuzana eliphilayo noma ikhompyutha ye-DNA eyinkimbinkimbi.
- I-Empirical Data Overestimation Igwenyiwe: Esikhundleni sokwengeza izibikezelo ezivela kumafomula ayisisekelo e-biophysical noma e-thermodynamic (okudonsa kanzima ukulandisa ngokuziphatha kwamangqamuzana okungaqondile), ithimba lakhe umtapo wolwazi 144 million amapheya alandelanayo ukuqeqesha i-BINND ngokuqondile kumicimbi engokoqobo, ebopha umhlaba wangempela.
- Ukunemba Okuphakeme Okubikezelayo: Ekuhloleni kobufakazi bomqondo, imodeli yokufunda ejulile ye-BINND izuziwe Ukunemba okungu-83.5%. ekubikezeleni ukuziphatha okubophayo, ukwenza kangcono kunamamodeli amanje wesimanje okungenani 10%.
- I-Asymmetrical Safe Failure: Lapho imodeli ye-AI yenza iphutha, yabonisa ukuchema okungabikezelwa kokuphepha: yayivame ukubikezela ukuthi imicu emibili ye-DNA izoba khona. hhayi bophezela lapho zenze khona ngempela, esikhundleni sokuqamba amanga ukuhlanganisa okungekho kuzokwenzeka. Lokhu kusiza abacwaningi bagweme ukuphazamiseka kwangemuva okuyinhlekelele (crosstalk) ekuxilongweni kwamangqamuzana.
- I-Matrix Demonstration: Ukuze kuboniswe insiza esebenzayo ye-BINND, ithimba lakhe isizindalwazi esisebenzisanayo esenza imephu ubudlelwano obuhlanganisayo bokulandelana kwezinhlamvu ezingamashumi ayisishiyagalolunye nesithupha ze-DNA ngokuqhathaniswa nezinye ezingamashumi amabili nesithupha ngokulandelana kwezinhlamvu ezingu-20, kusungulwa “incwadi yamakheli” ethembekile yokugcina kanye nokubuyisa idatha yamangqamuzana.
- Ukuvula i-Scalable DNA Computing: Ukugcina izingobo zomlando eziningi zedatha yomuntu ngaphakathi kwama-molecule e-DNA engabonakali kudinga ukubuyiswa kwedatha ebonakalayo esheshayo, engenaphutha. Ngokunikeza umgwaqo onokwethenjelwa wokuthi imiphi imicu ye-DNA eyonamathela ndawonye, i-BINND ixazulula inselele ebalulekile yokukala, ivula indlela yama-hard drive angakwazi ukugcina ama-petabyte edatha ethonjeni elilodwa.
Umthombo: I-North Carolina State University
Abacwaningi babonise imodeli ye-AI yenoveli engabikezela ukuthi yimaphi ama-molecule e-DNA abopha amanye ama-molecule e-DNA. Ukunikeza ukuqonda okuphelele kwalobu budlelwano obubophayo obuyi-hypercomplex buwusizo ezinhlelweni ezisukela kumathuluzi okuxilongwa kwe-biomedical kuya kukhompyutha ye-DNA.
“Sivame ukucabanga ngokubopha njengobudlelwane obulula kakhulu – I-Molecule A ibophezela ku-Molecule B,” kusho u-Albert Keung, umbhali ohambisana naye wocwaningo kanye noprofesa ohlangene wobunjiniyela bamakhemikhali kanye ne-biomolecular e-North Carolina State University. Kodwa ezimisweni zezinto eziphilayo, akulula neze.” I-molecule A ingase ibophe inqwaba yamanye ama-molecule, ngamazinga ahlukahlukene.
“Ukuthatha lokho ku-hypercomplexity kuyinselele enkulu, kodwa kubalulekile uma sifuna ukuqonda kangcono izinhlelo zofuzo zemvelo,” kusho uKeung, oyiGoodnight Distinguished Scholar in Innovation in Biotechnology and Biomolecular Engineering kanye nomqondisi woHlelo lwe-Biotechnology ku-NC State's Integrative Sciences Initiative.
“Futhi ukuthwebula lokho ku-hypercomplexity nakho kubalulekile uma sifuna ukuthuthukisa amathuluzi asebenzisa ngokugcwele ama-biomolecules, njengamathuluzi okuxilonga azwelayo umehluko wezakhi zofuzo noma izinhlelo zekhompyutha ze-DNA ezithembele ku-DNA ukuze igcine futhi ithole idatha.”
“Sasazi ukuthi amamodeli okufunda ajulile – amamodeli obuhlakani bokwenziwa akwazi ukuthatha amaphethini ayinkimbinkimbi – ayenamandla okusisiza sihlole lolu hlobo lwesistimu ye-hypercomplex,” kusho uGunavaran Brihadiswaran, umbhali ohamba naye wephepha kanye ne-Ph.D. umfundi e-NC State. “Kodwa-ke, besazi nokuthi sizodinga isethi yedatha eqinile ukuze siqeqeshe imodeli. Imodeli inhle kuphela njengedatha oyiqeqesha ngayo.”
Imizamo yangaphambilini yokuthuthukisa amathuluzi okubikezela ukuziphatha okubophezelayo kwe-DNA-DNA incike kumasethi edatha amancane uma kuqhathaniswa wedatha ye-DNA-DNA, yabe isisebenzisa amathuluzi okumodela e-biophysical ukubikezela ukuthi yikuphi ukulandelana kwe-DNA okungabophezela kukho okunye ukulandelana kwe-DNA. Amathuluzi abikezelayo angumphumela alwela ukuthwebula inkimbinkimbi yobudlelwano obubophezelayo.
“Sithathe indlela ehlukile yokuhlola eyasivumela ukuthi sikhiqize idatha eyengeziwe lapho ukulandelana kwe-DNA kuhlanganisa khona,” kusho uKarishma Matange, umbhali ohamba naye wephepha kanye ne-Ph.D. iziqu ze-NC State. “Sekukonke, imininingwane yethu egciniwe iqukethe amapheya alandelanayo ayizigidi eziyi-144. Le dathasethi ebanzi isivumele ukuthi sisebenzise amamodeli e-AI esikhundleni sokuphindaphinda ngokusekelwe ezimisweni ze-biophysical noma zamakhemikhali ezinto eziphilayo.”
Ngokukhethekile, abacwaningi basebenzise idathasethi yabo enkulu ukuqeqesha imodeli yokufunda ejulile ukubikezela ukuthi yikuphi ukulandelana kwe-DNA okuzobophezela kukho okunye ukulandelana kwe-DNA. Baqambe imodeli ethi BINND: Binding and Interaction Neural Network for DNA.
Ekuhlolweni kobufakazi bomqondo, abacwaningi bathola ukuthi imodeli ye-BINND yabikezela ukuthi yimaphi amapheya e-DNA azobopha ngokunemba okungu-83.5%. Futhi lapho yenza iphutha, yayivame ukubikezela ukuthi ukulandelana kwe-DNA okubili ngeke kubophe – kuyilapho ukulandelana ngabe bopha.
“I-BINND inembile okungenani i-10% kunemodeli yesimanje,” kusho uBrihadiswaran.
Ukukhombisa ukusetshenziswa kwe-BINND, abacwaningi basebenzise imodeli ukukhiqiza isizindalwazi esithwebula imvelo exhumene kakhulu yokuziphatha okubopha i-DNA-DNA. Isizindalwazi siyi-matrix, ebonisa ukuthi ukulandelana kwe-DNA yezinhlamvu ezingu-96 ezingu-20 kuhlanganisa kanjani – noma cha – nokunye ukulandelana kwe-DNA yezinhlamvu ezingu-26 ezingu-26.
“Lo mbukiso uwusizo lwangempela ngokombono wekhompiyutha ye-DNA, njengoba usinikeza ulwazi olubalulekile mayelana nezici zalokhu kulandelana – okubalulekile emizamweni yokuthwebula nokuthola ulwazi kusetshenziswa i-DNA,” kusho uJames Tuck, umbhali ohambisana naye wephepha kanye noprofesa wobunjiniyela bukagesi namakhompiyutha e-NC State. “Sithemba ukuthi abanye emphakathini wabacwaningi bazosebenzisa i-BINND, yingakho siyenza itholakale esidlangalaleni ku-GitHub.” Inqolobane ye-BINND ingatholakala kokuthi https://github.com/dna-storage/BINND.
“Enye yezinselelo zokugcinwa kwedatha ye-DNA kanye nekhompyutha kube ukuthi ingakhushulwa ukuze isetshenziswe ngendlela ebonakalayo,” kusho uKeung. “Sinethemba lokuthi i-BINND izoba yithuluzi elibalulekile lokwenza imizamo yokukhulisa lobo buchwepheshe, phakathi kwezinye izinhlelo zokusebenza ezingase zibe khona.”
Iphepha, “I-Deep Learning Predicts Dissimilar DNA-DNA Binding and Engineers Hyperconnected Networks,” ishicilelwe ukufinyelela okuvulekile kujenali. Ukuxhumana Kwemvelo. Leli phepha labhalwa ngokuhlanganyela nguKyle Tomek noKevin Volkel, bobabili abaPh.D. abathweswe iziqu ze-NC State; kanye no-Doug Townsend, onePh.D yamanje. umfundi e-NC State.
Uxhaso: Lo msebenzi wenziwa ngokusekelwa yiNational Science Foundation ngaphansi kwezibonelelo 2027655, 1901324 kanye 2403352; IziKhungo Zezempilo Zikazwelonke ngaphansi kwesibonelelo sika-R41HG013877; Usizo Lomnyango Wezemfundo Ezithweswe Iziqu Ezindaweni Ezidinga Ubudlelwano, P200A160061; kanye neSimons Foundation ngaphansi kwesibonelelo sika-990252.
Imibuzo Ebalulekile Iyaphendulwa:
A: Kubhayoloji yesikole samabanga aphezulu, sifunda ukuthi imicu ye-DNA ifana nezingcezu zephazili ezihlangana kuphela uma ziphikisana ngokuphelele. Kodwa ezinhlelweni zangempela zebhayoloji, amangqamuzana e-DNA anamandla kakhulu futhi anamathele. Umucu owodwa we-DNA ungase unamathele kwezinye izinto eziningi ezingaphelele, ubambelele ngamandla ahlukahlukene. Lokhu “hypercomplexity” kudala iwebhu enesiphithiphithi yokusebenzisana. Uma uzama ukwakha ikhompuyutha yebhayoloji, lokhu kunamathela okungalindelekile kungabangela izindlela ezihlukene zedatha ukuba ziwele futhi zehluleke, okwenza kube semqoka ukubikezela lokhu okufanayo okungaphelele kusengaphambili.
A: Isitoreji sedatha ye-DNA sisebenza ngokuhumusha i-digital 1s kanye ne-0s kuzisekelo ze-DNA zokwenziwa A, C, T, kanye no-G. Ukuze ubuyise ifayela elithile ku-molecular “hard drive,” ujova “i-probe” ye-fluorescent ye-DNA esebenza njengombuzo wosesho, ukuthola nokubophezela kuchungechunge lwedatha okuhlosiwe ukuze lufundeke. Uma uphenyo lwakho lokusesha lunamathela ngephutha kumafayela angalungile, idatha yakho iyonakala. I-BINND yenza njengomxhumanisi wezokuphepha ohlakaniphe ngendlela emangalisayo—enza imephu ngokunembile ukuthi lokhu kuhlelwa kuzohlanganyela kanjani ukuze kuqinisekiswe ukuthi imibuzo yakho yosesho ithola kuphela idatha elungile ngaphandle kwanoma iyiphi i-crosstalk yama-molecular.
A: Ekufundeni komshini, kunomthetho wegolide: imodeli inhle kuphela njengedatha esetshenziselwa ukuyiqeqesha. Imizamo yangaphambilini yokubikezela ukubophezela kwe-DNA incike kumasethi edatha amancane, okuphoqa amakhompyutha ukuthi enze ukuqagela okufundile ngokusekelwe kumafomula e-generic physics. Ithimba le-NC State lathatha indlela ehluke ngokuphelele. Ngokukhiqiza isizindalwazi esikhulu, somhlaba wangempela we 144 million amapheya alandelanayobanikeze imodeli yabo yokufunda ejulile umtapo wolwazi omkhulu wokusebenzelana kwangempela kwamangqamuzana ukuze kufundwe. Lesi sikali esikhulu sivumele i-BINND ukuthi ibone amaphethini abophayo ayinkimbinkimbi, afihliwe izibalo zefiziksi ezingenakukwazi ukubala.
Amanothi Omhleli:
- Lesi sihloko sihlelwe umhleli weNeuroscience News.
- Iphepha lejenali libuyekezwe ngokugcwele.
- Ingqikithi eyengeziwe yengezwe abasebenzi bethu.
Mayelana nale AI kanye nezindaba zocwaningo lwezofuzo
Umbhali: Matt Shipman
Umthombo: I-North Carolina State University
Othintana naye: Matt Shipman – North Carolina State University
Isithombe: Isithombe sifakwe ku-Neuroscience News
Ucwaningo lwangempela: Vula ukufinyelela.
“I-Deep Learning Predicts Dissimilar DNA-DNA Binding and Engineers Hyperconnected Networks” ka-Karishma Matange, Gunavaran Brihadiswaran, Kyle J. Tomek, Kevin Volkel, Doug Townsend, James M. Tuck & Albert J. Keung. Ukuxhumana Kwemvelo
I-DOI:10.1038/s41467-026-75395-w
Abstract
Ukufunda Okujulile Kubikezela Ukubopha Okungafani Kwe-DNA-DNA kanye Nonjiniyela Amanethiwekhi Axhumene Kakhulu
Izinhlaka ezivamile ku-molecular bioengineering kanye ne-synthetic biology igxila ku-orthogonality, ukubuka ukusebenzisana okubuthakathaka noma okungaqondile njengezinkinga okufanele zigwenywe. Lokhu kukhawulela isikhala sokulandelana esisebenzisekayo, kukhawula ukukala, futhi kushaye indiva izimo lapho amasistimu okwenziwa kufanele asebenze ngaphakathi kwesizinda semvelo sokuhlukahluka okulandelanayo okuphezulu. Ukubophela isikhala esigcwele kunzima ngenxa yokuthi amamodeli ayashoda akwazi ukubikezela ngokunembile nangokushesha ukusebenzisana okungewona ama-orthogonal futhi aqinisekiswe ngokumelene nedatha yeqiniso eliyisisekelo.
Lapha sithuthukisa i-BINND — I-Binding and Interaction Neural Network for DNA — sisebenzisa ukusebenzisana kwe-DNA-DNA njengendawo yokuhlola. I-BINND ihlanganisa inkundla yokuphuma ephezulu kakhulu ekala izigidi zokusebenzelana nemodeli yokufunda ejulile ethola ukunemba okungaphezu kuka-80%, eyenza okuvamile ngokulandelana okuhlukahlukene futhi isebenze ngokushesha izikhathi ezingu-50 kunamamodeli amanje.
Sibonisa ukubaluleka kwayo ngenethiwekhi eseshekayo ye-DNA yezinhlamvu zezinganekwane eziqanjiwe. I-BINND inika amandla ukubikezela okunembile kokuxilonga, i-bioengineering, ne-DNA origami, isekela ukushintshela ekusebenziseni isikhala esigcwele sokulandelana.
