Elix, CBI, 招待講演, ElixにおけるAI創薬と最新動向, 2021-10-26

Transcript

1. $PQZSJHIU˜&MJY *OD"MMSJHIUTSFTFSWFE
   $#*ֶձট଴ߨԋ
   &MJYʹ͓͚Δ"*૑ༀͱ࠷৽ಈ޲
   גࣜձࣾ&MJY
   $&0݁৓৳࠸
   2021/10/26
   1
   

   View Slide

2. $PQZSJHIU˜&MJY *OD"MMSJHIUTSFTFSWFE
   ૑ༀʹ͓͚Δ՝୊
   2
   Scannell et al. (2012)
   • 10೥Ҏ্ͷظؒɺ1000ԯԁҎ্
   ͷίετɺ੒ޭ཰ͷ௿͞
   

   • ૑ༀίετ͸ࢦ਺ؔ਺తʹ૿େ
   

   • 2010೥Ҏ߱͸͜ͷ૿Ճ͕ࢭ·ͬͨ
   ͱ͢Δݚڀ΋͋Δ΋ͷͷґવͱ͠
   ͯߴ͍
   Eroom's Law
   

   View Slide

3. $PQZSJHIU˜&MJY *OD"MMSJHIUTSFTFSWFE 3
   3FUIJOLJOH%SVH%JTDPWFSZ
   ૑ༀΛ࠶ߟ͢Δ
   

   View Slide

4. "*ؔ࿈ຽؒ౤ࢿֹ೥WT೥
   4
   https://aiindex.stanford.edu/report/
   2020೥Ͱ࠷΋େ͖͔ͬͨͷ͸૑ༀɻ͍ͭʹࣗಈӡసΑΓ΋େ͖ͳ౤ࢿֹʹɻ
   

   View Slide

5. $PQZSJHIU˜&MJY *OD"MMSJHIUTSFTFSWFE
   "*ελʔτΞοϓͱ੡ༀձࣾͷఏܞ਺
   5
   https://www.biopharmatrend.com/m/free-reports/ai/
   ߹ܭఏܞ਺͸ॱௐʹ૿Ճ
   

   View Slide

6. "*ར༻༻్ͷ಺༁
   6
   Schuhmancher et al. (2020)
   ௿෼ࢠ͕Ұ൪ଟ͍
   ※ ੈքͷ੡ༀձࣾτοϓ21͕ࣾ2014ʙ2019
   ʹ͔͚ͯऔΓ૊ΜͰ͍ͨAIϓϩδΣΫτ
   

   View Slide

7. 3FTUSJDUFE˜&MJY *OD
   ෼ࢠઃܭ
   7
   Image Source: Sanchez-Lengeling et al. (2018)
   Drug-likeͳ෼ࢠ͸ʙ1060ݸ
   ࣮ݧ/γϛϡϨʔγϣϯ ༧ଌϞσϧ ੜ੒Ϟσϧ
   

   View Slide

8. $PQZSJHIU˜&MJY *OD"MMSJHIUTSFTFSWFE
   &MJY%JTDPWFSZ5. "*%SVH%JTDPWFSZ1MBUGPSN
   
   8
   Elix͕ഓ͖ٕͬͯͨज़Λ݁ूͨ͠AI૑ༀϓϥοτϑΥʔϜ
   ᶃڞಉݚڀɹᶄϥΠηϯεఏڙ
   w ׆ੑ͋Γ
   w ׆ੑͳ͠
   w O.
   w ʜ
   

   View Slide

9. $PQZSJHIU˜&MJY *OD"MMSJHIUTSFTFSWFE
   &MJY%JTDPWFSZ5. "*%SVH%JTDPWFSZ1MBUGPSN
   
   9
   ࣗࣾݚڀΛؚΉ࠷৽ͷݚڀ੒ՌΛϓϥοτϑΥʔϜʹ࣮૷ɻࣗࣾݚڀͷྫΛަ͑ͭͭ঺հɻ
   ͜ͷޙͷεϙϯαʔηογϣϯͰ΋ϓϥοτϑΥʔϜΛ঺հ →
   ← CBIͰ΋5ͭޱ಄ൃද
   w ׆ੑ͋Γ
   w ׆ੑͳ͠
   w O.
   w ʜ
   

   View Slide

10. $PQZSJHIU˜&MJY *OD"MMSJHIUTSFTFSWFE
    &MJY1SFEJDUʢ༧ଌʣ
    10
    Properties
    Model
    • ׆ੑɺ෺ੑɺADMETͳͲΛ༧ଌ͢ΔϞδϡʔϧ
    
    
    • ैདྷϞσϧ͔Β࠷৽ͷάϥϑܥͷϞσϧ·Ͱ
    
    
    • ࣗಈͰϋΠύʔύϥϝʔλௐ੔Λͯ͠ϕετͳϞσϧΛબ୒
    Case Studyɿ
    

    Ξϯυϩήϯड༰ମʹର͢ΔόʔνϟϧεΫϦʔχϯά
    • ༧ଌ஋͚ͩͰͳ͘ɺCon
    fi
    denceείΞΛߟྀ͢Δ͜ͱͰ
    ΑΓ·ͱ΋ͳԽ߹෺Λબ୒
    
    
    • ߜΓࠐΜͩ53ݸͷ͏ͪ34ݸ͕ಛڐऔಘࡁΈͷ΋ͷͩͬͨ
    

    ʢ༧ଌϞσϧͷੑೳΛࣔ͢͜ͱ͕໨తͷ࣮ݧʣ
    Con
    fi
    denceείΞ͋Γ
    Con
    fi
    denceείΞͳ͠
    10/27 ޱ಄ൃද
    O4-3 Romeo Cozac
    

    “Graph Convolutional Networks for Ligand-based
    Virtual Screening against the Androgen Receptor”
    

    View Slide

11. $PQZSJHIU˜&MJY *OD"MMSJHIUTSFTFSWFE
    &MJY$SFBUFʢ෼ࢠੜ੒ʣ
    11
    Model
    • ੜ੒ϞσϧʹΑΓߏ଄ൃੜΛߦ͏Ϟδϡʔϧ
    
    
    • Elix Predictͱ૊Έ߹Θͤͯॴ๬ͷ׆ੑɾ෺ੑΛ࣋ͭ෼ࢠΛੜ੒
    
    
    • άϥϑϕʔεɺSMILESϕʔεɺϑϥάϝϯτϕʔεͳͲಠࣗϞσϧΛ
    ؚΉ༷ʑͳϞσϧΛαϙʔτ
    
    
    • ࢦఆͨ͠෦෼ߏ଄Λར༻͢Δ͜ͱ΋Մೳ
    eGEGLɿElixͷಠࣗϞσϧʢWüthrich et al. 2021ʣ
    
    
    • χϡʔϥϧωοτϫʔΫͱҨ఻తΞϧΰϦζϜΛ
    ߹ΘͤͨϞσϧ͕SOTAͱ͍͏എܠ
    
    
    • Ҩ఻తΞϧΰϦζϜ෦෼ʹυϝΠϯ஌ࣝΛೖΕΔ
    ͜ͱΛۃྗഉআͨ͠ΑΓҰൠԽͨ͠Ϟσϧ
    
    
    • ϕϯνϚʔΫͰಉ౳Ҏ্
    Wüthrich et al. (2021)
    10/27 ޱ಄ൃද
    O4-3 Pierre Wüthrich
    

    “Using Attribution-based Explainability to Guide
    Deep Molecular Optimization”
    

    View Slide

12. $PQZSJHIU˜&MJY *OD"MMSJHIUTSFTFSWFE
    3FUSPTZOUIFTJTʢٯ߹੒ղੳʣ
    12
    Model
    • ٯ߹੒ղੳΛߦ͏Ϟδϡʔϧ
    
    
    • େྔͷԽ߹෺Λ·ͱΊͯॲཧ΋Մೳ
    
    
    • Elix CreateͰੜ੒͞ΕͨԽ߹෺ͷ߹੒༰қੑͷݕ౼ʹ΋
    
    
    • ڞ௨͢ΔதؒମΛܦ༝͢Δϧʔτͷఏࣔ
    
    
    • Ձ֨ɺར༻Մೳͳࢼༀɺऩ཰౳Λߟྀͨ͠ϧʔτͷఏࣔ
    • ౦ژ޻ۀେֶͱͷڞಉݚڀ
    
    
    • ٯ߹੒ղੳπʔϧ܈
    
    
    • Φʔϓϯιʔεͱͯ͠ެ։༧ఆ
    Elix Synthesize
    10/27 ޱ಄ൃද
    O4-2 Haris Hasic
    

    

    “RetroSynthWAVE: An Open-
    Source Software Platform for
    E
    ffi
    cient Chemical Synthesis
    Research”
    

    View Slide

13. $PQZSJHIU˜&MJY *OD"MMSJHIUTSFTFSWFE
    σʔλ͕গͳ͍໰୊΁ͷରԠ
    13
    ࿈߹ֶशʢkMoL, Elix Milaʣ • ࿈߹ֶशɿσʔλΛ֎෦ʹग़͢͜ͱͳ͘ɺෳ਺اۀͷσʔλΛ׆
    ༻ֶͯ͠श͢Δ͜ͱΛՄೳʹ͢Δٕज़
    
    
    • kMoL: ࿈߹ֶशͱ༧ଌϞσϧͰߏ੒͞ΕΔϥΠϒϥϦ
    
    
    • ژ౎େֶͱڞಉ։ൃ
    
    
    • ࿈߹ֶशϞδϡʔϧElix MilaΛϕʔεʹ։ൃ
    
    
    • Φʔϓϯιʔεͱͯ͠ϦϦʔεࡁΈ (https://github.com/elix-tech/kmol)
    ࣗݾڭࢣ͋ΓֶशʢSelf-Supervised Learningʣ
    • ࣮ݧ஋͕ଘࡏ͠ͳ͍Խ߹෺σʔλ΋׆༻
    
    
    • ༧ଌਫ਼౓Λ޲্ͤ͞ΔΞϓϩʔν
    
    
    • σʔλ͕গͳ͍৔໘Ͱಛʹ༗༻
    10/27 ޱ಄ൃද
    O4-6 Laurent Dillard
    

    “Improving Molecular Property Prediction using
    Self-supervised Learning”
    

    View Slide

14. $PQZSJHIU˜&MJY *OD"MMSJHIUTSFTFSWFE
    ௿෼ࢠҎ֎ͷϞμϦςΟ
    14
    ௿෼ࢠ͕ϝΠϯͰ͋Δ΋ͷͷɺͦΕҎ֎ͷϞμϦςΟ΋ѻ͍ͬͯΔ
    ߅ମ H3ϧʔϓͷ3࣍ݩߏ଄༧ଌ
    • λϯύΫ࣭഑ྻσʔλʹΑΔڭࢣͳֶ͠शΛ׆༻
    
    
    • ͜ͷಛ௃நग़ʹΑΓH3ϧʔϓߏ଄༧ଌͷਫ਼౓Λ޲্
    10/27 ޱ಄ൃද
    O4-4 David Jimenez
    

    “Leveraging Self-Supervised Contextual Language
    Models for DNN Antibody CDR-H3 Loop Predictions”
    

    View Slide

15. $PQZSJHIU˜&MJY *OD"MMSJHIUTSFTFSWFE
    ϝϯόʔ
    15
    🌏 ੈքத͔Β༏लͳϝϯόʔΛ࠾༻
    
    
    • AIݚڀऀɺAIΤϯδχΞ
    
    
    • έϛετɺόΠΦϩδετ
    
    
    • ത࢜߸औಘऀଟ਺
    
    
    • ӳޠެ༻ޠ
    🇯🇵 🇭🇺 🇫🇷
    🇦🇺
    🇲🇾
    🇰🇬 🇺🇸
    🇳🇬 🇨🇭 🇧🇦 🇨🇴
    🏢 ૑ۀऀ
    ݁৓৳࠸ɺPh.D
    

    ڞಉ૑ۀऀɾCEO
    େٱอୡ໼
    
    
    ڞಉ૑ۀऀɾCOO
    ౦ژʹू·ͬͯ࢓ࣄ
    Λ͍ͯ͠·͢
    📍
    

    View Slide

16. Thank you!
    Contact us: info@elix-inc.com
    

    View Slide