Byagenze bite
Abashakashatsi berekanye MolEmb, urwego ruto rwo guhuza imiterere nini yindimi nini muburyo rusange bwo gushyiramo molekile. Sisitemu ihuza imyirondoro ya molekulari hamwe nibisobanuro byanditse mumwanya uhuriweho wo gushyiramo ukoresheje intego zinyuranye. Abanditsi batanga ibisubizo birushanwe kubijyanye no guhanura imitungo ya molekuline no gushyigikirwa na molekile-ihuza-inyandiko-yo kugarura. Bamenyekanisha kandi MolCAR, igipimo cyo kugerageza imiterere-imenyekanisha molekulari.
Uru rupapuro rwashyikirijwe arXiv ku ya 24 Kanama, rusobanura MolEmb nk'urwego rwo gusubiramo imiterere y’ururimi runini nka sisitemu yo gushyiramo molekile. Imiterere ya molekulari yerekana amakuru ajyanye na molekile mubice bishobora gukoreshwa na sisitemu yo hasi. Nk’uko abanditsi babivuga, kodegisi nyinshi zisanzwe ni sisitemu yinzobere yubatswe hafi ya molekile imwe kandi ikabyara ibice bitagabanijwe, bivuze ko guhagararirwa bidatandukanijwe neza nururimi-karemano rusobanura umurimo ugenewe cyangwa ibisobanuro bya molekile. MolEmb yashizweho kugirango ihuze imyirondoro ya molekile hamwe ninyandiko mumwanya uhuriweho.
Uburyo bwatangajwe bukoresha intego zinyuranye zigamije guhuza imyirondoro ya molekulari hamwe nibisobanuro byanditse. Mumagambo yagutse, imyitozo itandukanye ishishikarizwa guhuza molekulari hamwe ninyandiko zerekana kuba hafi mumwanya usangiwe mugihe utandukanya bibiri bidahuye. Inkomoko ntabwo itanga impapuro zirambuye zerekana imiterere, amakuru yamahugurwa, urutonde rwibanze, cyangwa ibisubizo byumubare, kubwibyo gushyira mubikorwa neza nubunini bwisuzumabumenyi byatangajwe ntibishobora gushingwa uhereye kubintu byatanzwe. Abanditsi baranga MolEmb nkuburemere, ariko isoko ntisobanura icyo kirego hamwe numubare wibara, kubara ibisabwa, cyangwa kugereranya nubundi buryo.
Abanditsi bavuga ubushobozi bubiri nyamukuru. Ubwa mbere, MolEmb isobanurwa nkirushanwa kubijyanye no guhanura imitungo ya molekile, icyiciro cyimirimo igereranya ibiranga molekile uhereye kubyo bahagarariye. Icya kabiri, ishyigikira kwambukiranya-modekile-y-inyandiko-yisubiramo mu mwanya umwe wo gushyiramo, ituma molekile nindimi byinjira bigereranywa binyuze mubisangiwe. Uru rupapuro rutangiza kandi MolCAR, isobanurwa nkigipimo cyo gusuzuma kugirango tumenye neza. Inkomoko ivuga ko ibipimo ngenderwaho byerekana ko gushyiramo imiterere ya molekulari byibanze ari umutungo wamakuru yubugenzuzi, bushimangira uburyo ingero zamahugurwa zubatswe aho gushingira gusa ku myubakire y’icyitegererezo.
Igikorwa cyatangiwe mu mahugurwa ya gatatu kuri Multi-modal Foundation Model hamwe na Moderi nini yindimi kubumenyi bwubuzima muri ICML 2026. arXiv igaragaza ko ari impapuro zamahugurwa zidafite ububiko kandi zikerekana urutonde rwatanzwe nkicyiciro cya mbere. Inkomoko yatanzwe ikubiyemo amakuru adasobanutse na bibliografiya, ariko nta cyemezo cyo gusuzuma-urungano, ububiko bwashyizwe mubikorwa, kwigana byigenga, amakuru yoherejwe, cyangwa ibimenyetso byerekana ko ubwo buryo bwatanze umukandida w’ibiyobyabwenge byemewe cyangwa byateje imbere ivuriro.
Ibisobanuro birambuye: arxiv.org ↗
Impamvu ari ngombwa
Ibikoresho bya molekulari nibisubirwamo byongeye bishobora gushyigikira imirimo nko guhanura umutungo, kwerekana ibintu, no kugarura. Niba imvugo yindimi nyinshi ishobora gutanga ibyapa bihinduka neza hamwe nururimi-karemano, abashakashatsi barashobora gukoresha urwego rumwe rworoshye rwo kugereranya imirimo myinshi yo kubara-chimie. Uru rupapuro rugaragaza kandi ko imyitwarire-ishingiye ku miterere iterwa ahanini namakuru yubugenzuzi, bigatuma igishushanyo mbonera cyerekana imibare ikomeye yubushobozi.
Kwinjiza molekulari nibikorwa remezo aho kuba umukoresha wa nyuma. Ikoreshwa rya vector ryongeye gukoreshwa rishobora kugaburira sisitemu nyinshi zo hasi, harimo guhanura imitungo, kwerekana ibintu, no kugarura. Ubusobanuro bushoboka bwa MolEmb rero nuburyo bwateganijwe guhinduka: icyitegererezo kimwe cya multimodal gishobora kugereranya amakuru ya molekile hamwe nibisobanuro byanditse kubyo umushakashatsi ashaka kugarura cyangwa guhanura. Urupapuro rufata nkuburyo bwo gukoresha kodegisi yihariye yihariye kubireba molekulari zitandukanye cyangwa imirimo.
Igishushanyo mbonera gishobora kuba ingirakamaro aho amakuru yubumenyi atandukanijwe hagati yimiterere ya molekulari yubusobanuro cyangwa ibisobanuro hamwe nururimi-karemano. Umwanya usangiwe urashobora koroha gushakisha molekile ukoresheje inyandiko cyangwa guhuza inyandiko za molekile hamwe nibisobanuro byanditse. Nyamara, inkomoko ishyiraho gusa ko abanditsi bavuga ko bashyigikiye molekile-ku-nyandiko. Ntabwo yerekana ko sisitemu isobanutse neza, ihendutse, yihuta, cyangwa yizewe kuruta ibikoresho byashizweho na chimie yihariye, kandi ntabwo yanditse inyandiko ikoreshwa nabashakashatsi hanze yisuzuma ryimpapuro.
Umwanzuro wimpapuro kubyerekeye amakuru yubugenzuzi ni ingenzi cyane mugusuzuma ibirego bijyanye na AI ya multimodal muri siyanse. Niba kumenyekanisha imiterere byigishijwe cyane cyane kuburorero bwakoreshejwe mukugenzura, noneho guhindura imiterere yicyitegererezo byonyine ntibishobora gukemura ibibazo nkibisobanuro bidasobanutse, ibisobanuro bituzuye, amagambo adahuye, cyangwa kubogama kubutaka bwa shimi. Ibigize imibare, ibimenyetso byerekana ibimenyetso, nubusabane hagati yimiterere yinyandiko namakuru ya molekuline bishobora kugira ingaruka mubisubizo. Inkomoko ntisobanura imibare yubugenzuzi yatanze ibisubizo byavuzwe cyangwa igereranya ingaruka.
Ingaruka ifatika iracyateganijwe. Abanditsi bavuga ko ibisubizo byabo byerekana ko imvugo nini y’ururimi runini ari inzira ifatika kandi yaguka yerekeza kuri moderi rusange yo gushyiramo molekile, ariko ibyo ni ubushakashatsi bwakozwe aho kuba ibimenyetso byerekana ko byemewe. Nta nkomoko ihari ivuga ko MolEmb yavumbuye ibiyobyabwenge, byateje imbere laboratoire, bisimbuza impuguke, cyangwa byinjijwe mu miyoboro ya chimie itanga umusaruro. Agaciro kako ako kanya nuburyo bwateganijwe bwo guhagararirwa hamwe nicyerekezo cyo gusuzuma siyanse ya siyanse ifashwa na AI.
Uburyo bukoreshwa: Uburyo bukora
Shakisha ikoranabuhanga ryihishe inyuma yiri terambere.
crm_get_transaction(id='4092').Which component of an AI application is the machine-learning model itself?
Ibyo kureba
Urupapuro nurupapuro rushya rwatanzwe, rutari archive impapuro zamahugurwa, kandi inkomoko ntizatanga ibisubizo byumubare, kugereranya, ingano ya dataset, guhuza kode, cyangwa ibimenyetso byerekana ko ukoresha ibiyobyabwenge. Igikorwa cyo gukurikirana kigomba gupima MolEmb kuri dataseti yagutse, indimi, kwerekana molekile, na laboratoire yigenga. Bizaba ngombwa kandi kumenya niba imiterere yimiterere itezimbere ibyemezo bya siyansi aho kuba amanota yo kugarura gusa.
Ikibazo cya mbere cyo gukurikirana ubushakashatsi ni ukumenya niba amarushanwa yatangajwe arushanwa afite imibare yigenga n'imirimo yigenga. Inkomoko ntabwo itanga amanota, igereranya ridashidikanywaho, ingano ya dataset, cyangwa izina ryibanze, bityo abasomyi ntibashobora kumenya ingano yatangajwe inyungu cyangwa ibibi. Isuzuma ry'ejo hazaza rigomba kugereranya MolEmb hamwe na kodegisi ya molekuline kabuhariwe hamwe na sisitemu ya multimodal mu buryo buhuye neza n'amahugurwa n'ibizamini.
MolCAR irashobora gufasha gusobanura niba sisitemu yumva imiterere cyangwa ihuza gusa amagambo asubirwamo hamwe nuburyo bwa molekile. Akamaro kayo kazaterwa nigipimo ngenderwaho gikwirakwizwa, ingorane, no kurwanya amakuru yamenetse. Inkomoko igaragaza MolCAR nk'igipimo cyo gusuzuma ariko ntisobanura ingero zayo, igishushanyo mbonera, ibipimo byo gusuzuma, cyangwa imiterere yo kurekura. Ibyo bisobanuro bizakenerwa mbere yuko igipimo gishobora gushyigikira imyanzuro yagutse kubyerekeye imiterere-yimikorere ya molekile.
Uruhare rwamakuru yubugenzuzi rusaba kandi gusuzumwa neza. Abashakashatsi bagomba gusuzuma niba imikorere ihinduka mugihe ibisobanuro bya molekuline bisobanuwe neza, bikozwe neza, byahinduwe, cyangwa bihujwe nibimenyetso bivuguruzanya. Bagomba kandi gusuzuma niba gushyiramo biguma bihamye mugihe inyandiko yanditswe idafite akamaro cyangwa idasobanutse. Nta na kimwe muri ibyo bizamini kivugwa mu isoko yatanzwe, bityo imipaka igena imiterere ikomeza kutamenyekana.
Hanyuma, kwemeza bifatika bisaba ibimenyetso birenze guhanura no kugarura ibipimo. Ikigereranyo cyumutungo wa molekuline hamwe nibisubizo byubushakashatsi birashobora kuyobora ubushakashatsi, ariko ntabwo ubwabyo byerekana ko molekile ifite umutekano, ikora neza, ikorwa, cyangwa ikwiriye gukoreshwa mubuvuzi. Ubushakashatsi bukurikirana bugomba gutanga raporo yibikorwa byororoka, ibisabwa byumutungo, ibibazo byatsinzwe, hamwe na laboratoire cyangwa isuzuma ryakazi aho bikenewe. Kugeza icyo gihe, MolEmb yunvikana neza nkurwego rwubushakashatsi butanga ikizere, ntabwo sisitemu yemewe yo kuvumbura ibiyobyabwenge.