Riba:
- Samun damar bambance inda a cikin tsarin kwayoyin halitta da jerin kwayoyin halitta (jeri, bambance-bambancen, tsari, omics, wallafe-wallafe) basirar wucin gadi yana adana ainihin lokaci da kuma inda yake da haɗari saboda ba shi da bayanan bayanai, bisa ga matakin hadarin aiki.
- Ikon gane tarko masu mutuwa guda uku kamar ƙirƙira jeri/gene/bambance-bambancen, rikice-rikice-version-nomenclature ruɗani da sifa na ƙarya, da kuma amfani da horo wanda ke tabbatar da kowane lamari na halitta a tushen farko.
- Ikon yin amfani da ka'idodin rashin sakin bayanan kwayoyin halitta na haƙuri a cikin nau'i mai ganewa don buɗe kayan aiki kuma koyaushe barin fassarar asibiti ta ƙarshe ga ƙwararren ƙwararren.
Halittar kwayoyin halitta da kwayoyin halitta shine kimiyyar karatu da fassarar abubuwa masu rai a cikin mafi girman harshensu - harshen DNA, RNA da sunadarai. A cikin wannan filin, kuskuren karanta harafi (biyu mai tushe), rikitar da sunan kwayar halitta, ko rarraba wani bambance-bambance (batu a cikin jerin kwayoyin halittar mutum wanda ya bambanta da abin da ake nufi); Yana iya haifar da ganewar asali ba daidai ba, magani mara amfani ko sakamakon binciken da ba daidai ba. Shi ya sa yin amfani da fasahar wucin gadi (AI) a wannan fanni na bukatar horo gwargwadon saurin gudu. A cikin wannan rukunin, za ku koyi inda kayan aikin da muke kira babban nau'in harshe (LLM - nau'in hankali na wucin gadi wanda ke samar da rubutu a ƙididdiga, tare da ma'anar "mafi dacewa kalmar") a zahiri adana lokaci a cikin ilimin halittar kwayoyin halitta da kwayoyin halitta, inda suke da haɗari, da kuma yadda za a tabbatar da kowane fitarwa.
Na farko, ra'ayi mai mahimmanci: hallucination shine lokacin da AI ke samar da bayanan da ba gaskiya ba ne, tare da cikakken tabbaci, kamar dai gaskiya ne. Wannan yana cikin kwayoyin halitta; Yana iya zuwa ta hanyar sunan jinsin da ba shi da shi, ƙirar bambance-bambancen lamba (misali "c.1234A>G") wanda ba ya wanzu, yanayin gado mara kyau, ko nuni ga labarin da ba shi da shi. Muhimmin mahimmin batu shine LLM ba shine bayanan kwayoyin halitta ko kayan aikin dakin gwaje-gwaje ba. Yana da janareta na rubutu wanda ke kwaikwayon kididdigar rubutun da yake gani a cikin bayanan horo. Yakan samar da ingantaccen ilmin halitta a mafi yawan lokuta saboda ya ga yawancin ingantattun rubutun ilmin halitta; amma bambanci tsakanin "gaskiya mafi yawan lokaci" da "gaskiya a kowane lokaci" na iya zama rayuwar mutum a cikin kwayoyin halitta.
A ina ne basirar wucin gadi ke aiki a wannan filin kuma a ina ba ya aiki?
Yi la'akari da AI azaman daftarin gaggawa, lamba, da abokin tarayya: mai mahimmanci amma mai mahimmanci don tabbatarwa. Bambanci mai zuwa shine kashin bayan wannan tsarin.
nema
Gudunmawar AI
Alhakin gwani
Binciken jeri
Yana rubuta lambar daidaitawa/bincike, yana fassara fitarwa
Gudun lambar kuma tabbatar da tsararru tare da bayanan tushen tushe
Bambancin fassarar
ACMG daftarin ma'auni yana ba da taƙaitaccen adabi
Tabbatar da kowace hujja a tushen asali, tabbatar da ajin
tsarin gina jiki
Yana Fassara fitowar AlphaFold, yana bayyana ƙimar amincewa
Tabbatar da ƙima mai aminci da kuma tabbataccen shaida
Tsarin CRISPR
Yana haifar da lissafin ɗan takarar gRNA da lamba
Tabbatar da kashe-manufa tare da kayan aikin ƙwararru
omics analysis
RNA-seq/ladin ƙididdiga yana ba da fassarar hanya
Tabbatar da ƙididdiga da ma'anar halitta
Adabi/rubutu
Yana yin taƙaitawa, daftarin aiki, gyare-gyaren harshe
Tabbatar da duk wani tunani da gaskiya a tushen
Dokokin yatsan hannu: Kada ka bari AI "tuna" bayanan kanta; amfani da shi don rubuta lambar, saita tsarin aiki, daftarin aiki da gyarawa; Koyaushe tabbatar da kowane gaskiyar ilimin halitta (jeri, bambance-bambancen, aikin kwayoyin halitta, akai-akai) daga tushe na asali abin dogaro. Madogarar farko anan ita ce madaidaitan bayanai kamar NCBI, Ensembl, UniProt, ClinVar, gnomAD, ko labaran da aka bita; Ba silsilar da LLM ke bayarwa daga zuciyarsa ba.
Tarkuna uku masu kisa na wannan filin
1. Jadawalin da aka yi, kwayoyin halitta da bambance-bambance. AI na iya "cika" jerin DNA wanda baya tunawa, bambance-bambancen daidaitawa, ko sunan kwayar halitta tare da wani abu mai kama da gaskiya. Ko da tsayin kirtani da haruffa sun bayyana daidai, ƙila ba za su dace da ainihin abin da ake nufi ba.
2. Haɗa kai da ruɗewar suna. AI; Siffofin Genome (GRCh37/hg19 zuwa GRCh38/hg38) na iya canzawa cikin shiru tsakanin tushen 0 da daidaitawa na tushen 1, wakilcin HGVS (tsarin rubutu na yau da kullun don bambance-bambancen karatu). Sakamakon ya zama "ma'ana" amma yana nuna matsayi mara kyau.
3. Halayen ƙarya da da'awar asibiti na ƙarya. AI na iya samar da labarin da aka yi gaba ɗaya a cikin jarida na ainihi, tare da sunayen marubuta masu sauti; ko kuma yana iya yin iƙirari ba tare da shaidar cewa bambance-bambancen shine “pathogenic” ba. Za mu yi cikakken bayani a cikin raka'a 8 da 9.
Tukwici: Kusanci kowace fitowar AI ta halitta tare da tambayoyi uku: (1) Wane tushe na farko ya tabbatar da wannan gaskiyar (jeri, bambance-bambancen, gene)? (2) Shin sigar genome da tsarin daidaitawa daidai ne? (3) Ta yaya zan tabbatar da wannan da kaina? Waɗannan tambayoyin guda uku suna kama mafi yawan kurakurai a cikin toho.
Mataki-mataki: amintaccen aikin aiki na AI
1. Ƙayyade aikin tare da mahallin da siga. "Me wannan kwayar halittar take yi?" maimakon a zahiri rubuta mahallin, kwafi, da sigar genome, kamar "Ina so in kimanta tasirin aikin bambance-bambancen mai zuwa a cikin sigar GRCh38, kwafin NM_007294.4 na gene BRCA1."
2. Bukatar tushe da tabbatarwa. Tambayi AI don tantance wace bayanan bayanai don bincika kowace hujja. Umurnin "Idan ba ku sani ba, kada ku gyara shi, ku ce 'dole ne a tabbatar da shi'" yana rage hasashe amma bai ƙare ba.
3. Tabbatar da lambar ta gudu ko amfani da kayan aiki. Tabbatar da ayyukan tsararru tare da lambar Python (Biopython) mai aiwatarwa, bambance-bambancen daidaitawa tare da mai canzawa na yau da kullun, tsari tare da makin bayanai na AlphaFold.
4. Yi gwajin nazarin halittu. Shin firam ɗin codon yana riƙe? Shin yawan adadin bambance-bambancen (gnomAD) ya dace da cutar? An shafa yankin sunadaran? Waɗannan suna kama kurakuran da AI ke ɓacewa.
5. Yi bincike na sirri da ɗabi'a. Kada a taɓa manna bayanan kwayoyin halitta na haƙuri cikin kayan aikin AI na jama'a a cikin sigar da za a iya ganewa. Za mu rufe wannan dalla-dalla a cikin raka'a 10.
uku mini lokuta
Case 1 - Bambance-bambancen da aka yi. Wani mashawarcin kwayoyin halitta ya tambayi AI ma'anar "CFTR c.1521_1523delCTT" a cikin rahoton majiyyaci kuma ya sami cikakken bayani. Duk da haka, yayin da YZ kuma ya rubuta wannan tare da wani rubutu na daban kamar "p.Phe508del", ya kara da wani abin da aka yi "c.1600A>T" a wata tambaya, kodayake ya ba da wurin bambance-bambancen daidai. Lokacin da mashawarcin ya bincika ClinVar, ya ga cewa bambancin na biyu ba ya samuwa kwata-kwata. Lokacin da aka rasa: minti 4; An hana kuskure: shigar da bambance-bambancen karya a cikin rahoton.
Case 2 - Haɗa tarko. Wani mai bincike ya tambayi AI don daidaitawar kwayoyin halitta. AI ta ba da haɗin gwiwar hg19, amma aikin mai binciken yana amfani da hg38. Haɗin kai ya koma kusan sansanonin 3,000. Mai binciken ya lura da bambanci lokacin da ya duba hoton tare da kayan aikin "lifOver" (inter-version coordinate transformation). Ba tare da tabbatarwa ba, gabaɗayan jeri zai yi kuskure.
Case 3 - Tabbatar da samu. Wani ɗalibin da ya kammala karatun digiri ya tambayi AI don lambar Python wanda ke gano buɗaɗɗen firam ɗin karantawa (ORF - yankin furotin-coding) na jeri. Lambar ta yi aiki, amma ta sami gajeriyar furotin saboda tana neman fara codon a cikin firam ɗin da ba daidai ba. Lokacin da ɗalibin ya kwatanta sakamakon tare da sanannen sunadarin magana, ya lura da bambancin tsayi, ya nemi AI ta bincika dukkan firam ɗin uku, kuma ya gyara shi cikin mintuna 10.
Samfura huɗu masu kwafi
1) Tushen da ake buƙata, tabbataccen fassarar:
Matsayinku: Mataimakin kwayoyin halitta. Kimanta gaskiyar mai zuwa: [gene/variant/sequence]. A bayyane sigar genome (GRCh37/38) da kwafi. Ga kowane da'awar, rubuta a cikin wane tushe na farko (NCBI, Ensembl, UniProt, ClinVar, gnomAD) yakamata a tabbatar dashi. KADA KA samar da kowane jerin / daidaitawa / bambance-bambancen da ba ka da tabbas game da shi; Rubuta "dole ne a tabbatar".
2) Tabbatar da aikin tsararru tare da lambar:
Rubuta lambar Python (Biopython) mai aiwatarwa wanda ke yin nazarin kirtani mai zuwa: [aiki].Lambar; Bayyana sigar genome, firam, da zato a cikin layin sharhi. Ƙara matakin tabbatarwa don kwatanta sakamakon zuwa sanannen tunani.
3) Fara jera kasada:
Kafin yin wannan aikin kwayoyin halitta, jera kurakurai guda 5 da aka fi sani a wannan aikin da yadda ake guje wa kowane: [aiki]. Mayar da hankali musamman akan tsarin daidaitawa, sigar genome, da kurakuran nomenclature.
4) Ikon sirri:
Kafin ba da wannan rubutun ga kayan aikin AI, wane bayani ya ƙunshi wanda zai iya gano majiyyaci (suna, lambar ID, kwanan wata, haɗuwa da ba kasafai ba)? Ta yaya zan iya ɓoye sunayen waɗannan? Ba da shi a cikin lissafi.
Rauni mai ƙarfi / Ƙarfi mai ƙarfi
Rauni: "Shin wannan maye gurbi a cikin BRCA1 yana cutarwa?"
Matsala: Babu sigar genome, babu kwafi, bambance-bambancen zayyana, ba a nemi tushe ba. AI na iya yin fassarar saman kai.
Ƙarfafa: "Ina so in kimanta bambancin NM_007294.4: c.68_69delAG a cikin rubutun GRCh38, BRCA1 (NM_007294.4) bisa ga ka'idojin ACMG. Fada mani abin da zan nema a ClinVar da gnomAD ga kowane yanki na shaida; Kada ku jera ainihin abin da ake bukata na bayanan.
Me ya sa yake da ƙarfi: Share mahallin, sigar, kwafi, daidaitaccen bayanin kula, da hanyar tabbatarwa; An takaita fagen ƙirƙira AI.
Kuskuren gama gari
- Ba tantance sigar genome ba. Gudanar da daidaitawa tsakanin hg19 da hg38; Duk haɗin kai da aka bayar ba tare da siga ba yana da shakku.
- Kai tsaye ta amfani da jeri/bambancin da AI ke bayarwa. Kowane jeri da bambance-bambancen dole ne a tabbatar da su a tushen farko.
- Barin shawarar asibiti zuwa AI. AI na iya "gaya" aji mai cutarwa, amma fassarar asibiti ta ƙarshe ta ta'allaka ne da ƙwararren ƙwararren.
- Manna bayanan haƙuri cikin buɗaɗɗen kayan aikin. Bayanan kwayoyin halitta da za a iya gane su sun ƙunshi keta sirrin sirri.
- Ruɗewar nomenclature. c., p., g. Haɗin wakilci da nau'ikan kwafi suna nuna bambance-bambancen da ba daidai ba.
Tsanaki: Sautin "aminci" na AI ba shaida ce ta daidaito ba. Mafi hatsarin hasashe sun zo tare da mafi kyawun magana da gamsarwa. Lokacin da kuka ji sautin ƙarfin hali, buƙatar tabbatarwa tana ƙaruwa, ba ta raguwa ba.
A takaice
- AI a cikin ilmin kwayoyin halitta da kwayoyin halitta; Mataimaki ne mai ƙarfi wanda ke rubutawa, zayyanawa, sharhi, da kuma gyara lambar - amma ba bayanan bayanai ba ne ko kayan aikin lab.
- Tarko masu kisa guda uku: jeri mai ban tsoro/gene/bambance-bambancen, rikice-rikice-version-nomenclature, da'awar asibiti na ƙarya.
- Dole ne a tabbatar da kowace hujja ta ilimin halitta a tushen farko; Dole ne a tabbatar da kowace lambar ta gudanar da ita.
- Ƙarshen alhakin asibiti da kimiyya, gami da sirri da ɗa'a, koyaushe yana tare da ƙwararren ƙwararren.
Aikin aikace-aikace
Don jinsi da bambance-bambancen kuna da (ko samfurin), tambayi AI don kimantawa ta amfani da samfuri 1 a sama. Sannan da kanka bincika kowace hujja AI ta dawo ( sigar genome, kwafi, wakilcin bambancin) a cikin ClinVar da gnomAD. Yi ƙoƙarin nemo bambanci tsakanin AI da tushen aƙalla aya ɗaya kuma ku lura da wannan bambanci a cikin jumla ɗaya.
jerin abubuwan dubawa
- [ ] Na bayyana aikin ta hanyar sigar genome, kwafi, da mahallin mahallin.
- [ ] Na tambayi AI don tushen farko don kowace hujja.
- [ ] Ni da kaina na tabbatar da kowane jeri/daidaitawa/ bambanta.
- [ ] Na tabbatar ta hanyar gudanar da lambar ko tare da kayan aiki.
- [ ] Na duba sirrin bayanan mara lafiya.
- [ ] Na amince da sharhin ƙarshe da kaina, Ban bar shi ga AI ba.