Raka'a
1. Gabatarwa zuwa Haƙiƙa na Artificial a Chemistry: Matsayi, Iyakoki, Tabbatarwa da Da'a 2. Wakilin Kwayoyin Halitta da Tsarin Sinadarai: Tabbatarwa tare da SMILES, InChi da RDKit 3. Hasashen amsawa da injina: Samfur, yanayi da Hasashen Hasashen Mummuna 4. Taimakon Fassarar Bakan: NMR, IR da Mass Spectrometry 5. Bita na wallafe-wallafe da Tsare-tsaren Haɗa: Tushen, Tsari da Taswirar Hanya 6. Binciken Bayanai na Chemical da Python: pandas, Stoichiometry da Calibration 7. Hasashen Dukiyar Kwayoyin Halitta da QSAR: Ƙaddamarwa, pKa da Iyakan Model 8. Takardun dakin gwaje-gwaje: Littafin Bayanin Gwaji, ELN da Maimaituwa 9. Ƙimar Tsaro da Haɗari: GHS, SDS da Iyaka na Hankali na Artificial 10. Tabbatarwa, Hallucination, da Mutuncin Kimiyya 11. Dan Adam a Gwaji: Da'a, Keɓantawa, Nauyi da Ƙarshe-zuwa Ƙarshen Aiki
Naúrar 1 / 11

Gabatarwa zuwa Haƙiƙa na Artificial a Chemistry: Matsayi, Iyakoki, Tabbatarwa da Da'a

Riba:

  • Samun damar bambance inda hankali na wucin gadi ke adana lokaci a cikin aikin sinadarai (ra'ayi, ƙira, bincike, bayar da rahoto) da kuma inda aka bar tsarin, lamba, albarkatu da yanke shawara na aminci ga mutane, dangane da matakin haɗari.
  • Ikon yin amfani da horo wanda ke bincika kowane fitarwa da kansa tare da matakai huɗu na Tsarin-Lambar-Haɓaka-Tsaro.
  • Fahimtar dalilin da yasa abubuwan da aka yi da lambobi, nassoshi na ƙarya, da kuskuren tsaro sune haɗari ga wannan sana'a wanda dole ne a yi la'akari da shi tun daga farkon.

Chemistry kimiyyar gwaji ce da ke nazarin tsari, canji da kaddarorin kwayoyin halitta. Masanin sunadarai yana auna, narkar da, zafi da ma'auni a cikin dakin gwaje-gwaje; A teburinsa, yana zana kwayoyin halitta, yana tsara halayen halayen, yana fassara spectra, bincika wallafe-wallafen kuma yana rubuta rahotanni. Hankali na wucin gadi (AI a takaice, anan musamman software wanda ke samar da rubutu ta hanyar tsinkayar “mafi yiwuwa kalma ta gaba,” wanda aka horar akan babban tsarin harshe, watau, manyan bayanan rubutu) na iya saurin sauri da yawa daga cikin wadannan ayyukan tebur. Amma ba ya gina dakin binciken da kansa, ba ya karanta ma'auni, kuma mafi mahimmanci: baya tabbatar da kowane lambobi, ƙididdiga, ko ƙididdiga da yake samarwa a madadin ku. Wannan tsarin yana koyar da yadda ake amfani da AI a cikin sinadarai daga ƙarshe zuwa ƙarshe amma cikin kulawa.

A cikin wannan rukunin farko, ina so in dasa ra'ayi ɗaya a cikin zuciyar ku: AI mataimaki ne a cikin ilmin sunadarai, ba masanin sinadarai ba. Yana rubuta lamba, yana samar da faci, gina tsarin retrosynthesis, yana taimakawa fassara kololuwar bakan, taƙaita wallafe-wallafe. Duk da haka, ƙwararren masanin ilmin sinadarai ne ya yanke shawarar ko za'a iya haɗa kwayoyin halitta, ko amsawar ba ta da lafiya ko a'a, ko nassoshi na gaske ne ko a'a, da kuma babban alhakin.

A ina AI ya dace da aikin sinadarai?

Bari mu karkasa aikin sinadari dalla-dalla zuwa matakai biyar: (1) ra'ayi da wallafe-wallafe, (2) ƙirar ƙwayoyin cuta da ƙirar amsawa, (3) aikace-aikacen dakin gwaje-gwaje, (4) nazari da haɓakawa, (5) bayar da rahoto. AI yana ɗaukar ƙima daban-daban a waɗannan matakan.

  • Ra'ayoyi da wallafe-wallafe: AI da sauri ya taƙaita wani batu, ya bayyana mahimman ra'ayoyi, yana haifar da kalmomin bincike. Amma yana iya yin sifa; Za mu ga wannan nan da nan.
  • Zane: Yana Ƙirƙirar SMILES (nau'i na rubuta kwayoyin halitta azaman sigar rubutu; ƙari akan wancan daga baya), yana ba da shawarar matakan sake dawowa, jera hanyoyin amsawa. Shawara tana da kyau, ba garanti ba.
  • Laboratory: AI ba ya yin komai a nan. Mutane ne ke aiwatar da matakan auna nauyi, dumama da tsaro. AI kawai yana taimakawa rubuta hanyar; Kafin aiwatar da wannan hanya dole ne mutum ya kula da ita don aminci.
  • Analysis: A cikin fassarar NMR, IR, ma'auni mai yawa (waɗannan dabarun aunawa ne waɗanda ke nuna tsarin kwayoyin halitta), YZ yana taimakawa wajen tara kololuwa da jera abubuwan da za a iya. Aikin ƙarshe na ku ne.
  • Rahoto: Yana samar da littafin rubutu na gwaji, daftarin labari, daftarin tsari na aminci. Yana da ƙarfi sosai; amma yakamata lambobin suyi daidai da naku.
Tukwici: Mafi aminci wurin amfani da AI yana cikin "idan wani abu ya yi kuskure za a lura da shi kuma a gyara shi nan da nan" yankuna: daftarin rubutu, kwarangwal na lamba, bayanin lokaci. Wurare masu haɗari sune "idan ƙarya, yana yaduwa cikin shiru" yankunan: madaidaicin, nassoshi, tabbacin tsaro, sakamakon lambobi.

Me ya sa ya kamata mu yi hankali? Hatsari guda uku na tsari

Babban samfurin harshe ba ƙididdiga ba ne ko injin sinadarai. Yana fitar da rubutu wanda a kididdiga "da alama". Wannan yana haifar da haɗari guda uku a cikin sinadarai:

  1. Ƙirƙira (hallucination): Za a iya amincewa da rubuta wani fili wanda ba abin ƙira ba, amsawar da ba ta wanzu, ko nauyin kwayoyin da ba daidai ba. Misali, ya ce "Nauyin kwayoyin halitta na fili X shine 154.2 g/mol", yayin da madaidaicin ƙimar shine 168.2.
  2. Ƙirar karya: Yana iya ba da tushen da yake kama da gaskiya amma babu shi kwata-kwata, kamar "Smith et al., 2019, Journal of Organic Chemistry." Har ma yana iya yin lambar DOI.
  3. Rashin aminci: Zai iya gabatar da haɗin haɗari na reagents (misali oxidizer da sauran kaushi na halitta a cikin rabon da bai dace ba) azaman "babu matsala".

Waɗannan hatsarori guda uku za su bayyana akai-akai a kowace raka'a na wannan tsarin. Maganin ba shine kubuta daga AI ba, amma don kafa tsarin tabbatar da kowane fitarwa tare da tushe mai zaman kansa.

kananan lokuta

Case 1 - Daidaitaccen nauyin kwayoyin halitta. Wani dalibi da ya kammala karatun digiri ya tambayi AI nauyin kwayoyin para-nitrophenol. "139.11 g/mol," in ji AI. Dalibin ya kasance mai ƙarfin gwiwa kuma ya shirya yin nauyin 6.96 g don 0.05 mol. Koyaya, nauyin kwayoyin para-nitrophenol shine 139.11 g / mol, kuma wannan lokacin daidai ne - amma ɗalibin bai taɓa bincika ba. A kan fili na gaba (para-aminophenol, ainihin ƙimar 109.13 g / mol) YZ ya ce "123.15"; Wannan lokacin ɗalibin ya kama shi ta hanyar ƙididdige shi da hannu daga dabarar (C6H7NO2): 6 × 12.011 + 7 × 1.008 + 14.007 + 2 × 15.999 = 109.13. Darasi: lissafta kowane nauyin kwayoyin halitta ba tare da dabara ba.

Case na 2 - DOI na karya. Wani mai bincike ya tambayi AI don tushen 5 a cikin nazarin wallafe-wallafensa. DOI na biyu daga cikin biyar sun dawo "ba a same su ba" lokacin da aka danna; Kanun labarai na gaskiya ne, amma labaran ba su kasance ba. ɓata lokaci minti 40. Darasi: Ba kowace magana ce za a yi amfani da ita ba tare da tabbatarwa a cikin bayanan (DOI, PubMed, Reaxys).

Case na 3 - Hazard da alama lafiya. Wani mai amfani ya tambayi AI game da hanyar tsaftacewa; AI a kaikaice ya samar da shawarar haɗa wani bayani mai ɗauke da hypochlorite tare da maganin acidic-haɗin da zai iya sakin iskar chlorine. An yi sa'a mai amfani ya kalli takardar bayanan aminci (SDS) ya tsaya. Darasi: kowace hanya ana bincikar ɗan adam don SDS da haɗari kafin a aiwatar da su.

Rauni mai ƙarfi / Ƙarfi mai ƙarfi

Rawanin faɗakarwa:

Haɗa wannan kwayar halitta.

Wannan da'awar ba ta da tabbas: menene kwayoyin halitta, menene kayan farawa, menene ma'auni, menene ƙuntatawa? AI yana daidaita gibba.

Ƙarfi mai ƙarfi:

Matsayi: Kai gogaggen mataimaki ne a cikin haɗaɗɗun kwayoyin halitta. Kwayoyin manufa: 4-methoxyacetophenone (SMILES: COc1ccc (cc1) C (C) = O). Muna da anisole azaman kayan farawa. Aiki: Ba da shawara na retrosynthesis mataki 2. Ga kowane mataki: reagents, yanayi (zazzabi, sauran ƙarfi), da POSSIBLE gefen halayen. Ƙuntatawa: Alama "TABBATA" inda ba ku da tabbas, dacewa.Fitarwa: matakai masu lamba + jerin maki don tabbatarwa.

Bambanci: rawar, ainihin maƙasudi (tare da SMILES), mahallin, adadin matakai, tsarin fitarwa da ƙuntatawa "daidai". Waɗannan su ne ɓangarorin biyar na ingantacciyar ƙarfafawa a cikin ilmin sunadarai: rawar, ainihin wakilcin tsari, mahallin, ɗawainiya, ƙuntataccen tabbaci.

Nau'in kayan aikin AI: samfurin harshe ko kayan aikin sunadarai?

A cikin ilmin sunadarai kun haɗu da "AI" guda biyu daban-daban kuma yana da mahimmanci kada ku dame su:

Salon

Me yake aikatawa

dogara

Misali amfani

Samfurin harshe na gaba ɗaya

Yana haifar da rubutu/ladi, yayi bayani, rubuta daftarin aiki

Ƙananan lambobi, babba akan harshe

Tsarin tsari, bayanin kalmomi

Laburaren Chemistry (RDKit da dai sauransu)

Yana aiwatar da kwayoyin halitta deterministically, yana ƙididdige nauyin kwayoyin halitta

Babban (bisa ka'ida)

Tabbatar da SMILES, MA asusu

Samfurin tsinkayar al'ada (retrosynthesis, fasali)

Yana ba da hasashen da aka horar da bayanai

Matsakaici, ya danganta da yanki

Retrosynthesis, kimanta solubility

Kayan aikin adabi/bincike

Tambayoyin bayanai na ainihi

Babban dangane da tushe

Tabbacin ra'ayi, neman amsa

Mafi ƙarfin aiki mai ƙarfi ya haɗa waɗannan: ƙirar harshe yana haifar da ra'ayin, ɗakin karatu na ilmin sunadarai yana ƙididdige adadin ƙididdigewa, kayan aikin adabi yana tabbatar da sifa, ɗan adam ya inganta. Za mu kafa wannan sarkar a raka'a ta gaba.

Horon tabbatarwa: Matakai HUDU

Ɗauki kowane fitowar AI ta waɗannan matakai huɗu kafin shigar da shi cikin lab ko rahoto:

  1. Tsarin: Shin bayanin kula da kwayoyin halitta/masu amsa (SMILES/InChi) yana aiki? Za a iya mirgine shi da kayan aiki (RDKit)?
  2. Lamba: Shin an tantance nauyin kwayoyin halitta, stoichiometry, lissafin yawan amfanin ƙasa da kansa (da hannu ko ɗakin karatu)?
  3. Madogara: Shin an tabbatar da kowace iƙirari da sifa a cikin ainihin ma'ajin bayanai?
  4. Tsaro: Shin an karanta SDS ga kowane reagent a cikin hanyar, shin akwai haɗuwa masu haɗari?
Lura: Waɗannan matakai guda huɗu suna aiki "ko da yaushe", ba "wani lokaci ba". Yanayin 20 wanda AI daidai yake ba sa uzuri yanayin 1 wanda ba daidai ba; saboda a cikin ilmin sunadarai cewa yanayi 1 na iya zama fashewa, guba ko kuma labarin da aka janye.

Kuskuren gama gari

  • Kuskure fitowar AI a matsayin "albarkatu". AI ba hanya ba ce, amma janareta wanda ke ƙoƙarin tunawa da albarkatu (wani lokaci ba daidai ba). Ita ce tushen bayanai.
  • Amincewa da lambar. Amfani da lambobi kamar nauyin kwayoyin halitta, pKa, wurin tafasa ba tare da tabbatar da su ba. Ana iya ƙididdige waɗannan / bincika ƙayyadaddun ƙididdiga; Tambayar samfurin harshe shine hanya mafi rauni.
  • Fitar da tsaro zuwa AI. "AI bai ce yana da haɗari ba" ba yana nufin yana da lafiya ba. Ƙimar haɗari shine aikin mutum da SDS.
  • Tashin hankali. Ba ba da kwayoyin halitta da sunansa amma ta tsarinsa (SMILES/InChi); Wannan yana haifar da kwayoyin da ba daidai ba saboda rikicewar suna.
  • Amincewa da gwadawa ɗaya kawai. sake yin tambaya iri ɗaya ta wata hanya dabam da rashin bincika daidaito.

A takaice

  • AI shine mataimaki na tebur mai ƙarfi a cikin ilmin sunadarai: yana samar da lamba, ƙayyadaddun bayanai, kwatancen, skeleton scan; amma dakin gwaje-gwaje baya yinsa kuma baya tabbatar da fitowar sa.
  • Akwai hatsarori guda uku: fili/lamba na karya, sifa ta karya, rugujewar tsaro.
  • Ware da haɗa samfurin harshe na gaba ɗaya da kayan aikin sinadarai masu kayyade (RDKit, bayanan bayanai).
  • Ɗauki kowane fitarwa ta matakai huɗu na Tsarin-Lamba-Source-Tsaro.
  • Kyakkyawan faɗakarwa: ya haɗa da rawar, bayyanannen wakilcin tsari, mahallin, ɗawainiya, ƙaƙƙarfan tabbatarwa.

Aikin aikace-aikace

Zaɓi kwayoyin da kuka yi nazari da kanku (misali aspirin, SMILES: CC(=O)Oc1ccccc1C(=O)O). Tambayi AI abubuwa uku: (1) nauyin kwayoyin halitta, (2) takaddun magana guda biyu, (3) matakin haɗin gwiwa. Sannan tabbatar da kowane da kansa: da hannu kirga nauyin kwayoyin halitta daga dabarar, tabbatar da ambato tare da DOI, bincika matakin haɗin gwiwa tare da ido na tsaro. Wanne fitarwa ya fito daidai kuma wanne ya buƙaci gyara? Ajiye rabin shafi "takardar tabbatarwa."

jerin abubuwan dubawa

  • [ ] Na gane cewa AI mataimaki ne, ba masanin kimiyya ba.
  • [ ] Na gane haɗarin ƙirƙira, sifa na ƙarya, da rashin yarda da misalai.
  • [ ] Zan iya bambance tsakanin ƙirar harshe da kayan aikin sinadarai masu ƙima.
  • [ ] Zan yi amfani da matakai huɗu na Tsarin-Lamba-Source-Tsaro ga kowane fitarwa.
  • [ ] Na kwatanta kwayoyin halitta ta hanyar bayanin tsari (SMILES/InChi), ba da suna ba.
  • [ ] Na ajiye aƙalla rajistan tabbatarwa ɗaya.
Na gaba