Ce diaporama a bien été signalé.
Nous utilisons votre profil LinkedIn et vos données d’activité pour vous proposer des publicités personnalisées et pertinentes. Vous pouvez changer vos préférences de publicités à tout moment.
Chemical Biology Laboratory
Center for Cancer Research
National Cancer Institute
National Institutes of Health
Synthetical...
Idea of SAVI
2
What can I
make
easily,
reliably,
safely
and
cheaply?
Not in screening databases
Many steps and very
costl...
Goal of SAVI
3
Reactant
A
Reactant
B
SAVI
Product
you can buy the
starting materials
1-step
reaction
no exotic
equipment
o...
SAVI components
4
CACTVS
Lhasa Ltd (UK), LHASA LLC (US)
Building Blocks Transforms
Chemoinformatics
engine
LHASA transform...
Reaction enumeration tools & libraries based on them
 REAL database – Enamine
• 145k building blocks X 120 transforms
• R...
CHMTRN/PATRAN transform example
6
DISCONNECTIVE
SUBGOAL*ALLOWED
BROKEN*BONDS BOND*5 BOND*8 BOND*1
...
IF CARBON ON ALPHA T...
CHMTRN/PATRAN transforms: smarter than SMARTS
7
24
14
IF BOND*6 IS A FUSION BOND
THEN SUBTRACT 10
34
ADD 10 IF ARYL ON ATO...
Workflow
8
Building
blocks Pattern matching
building blocks for each
transform
Transforms
reactant1
building blocks
reacta...
First SAVI production run (2016)
9
• CHMTRN/PATRAN transform logic
disqualified 17% of reactions
• Unique products
Downloa...
10
Product property distributions
0
5
10
15
20
100
200
300
400
500
600
700
800
900
1000
>1000
Millions
Mol Wt.
Molecular w...
SAVI produces novel ring systems
Total novel SA rings/scaffolds created: 999
(1) Ertl et al., J. Med. Chem. 49, 4568–4573 ...
12
Annotations for drug design
 “Drug design” properties:
• RO5 violations, PAINS matches, fsp3, log P…
• All 275 Bruns &...
 ~600k building blocks
 ~20 chemistries (~50 transforms) implemented
• 7 entirely new chemistries
• example of new chemi...
14
1 M
10 M
100 M
1 B
10 B
100 B
100 K
10 K
1 K
~100 Billion total possible pairs
Possible reactant pairs: >> 1 billion
15
 To generate comparable number of products for each transform
 Select 1 billion truly “interesting” compounds from 10...
More and better annotations planned
16
Predicted reaction conditions
Determine if SAVI product is a mixture
Annotate with
...
Ongoing work
17
grow molecule
elsewhere
Preserve reactant’s
functional groups already
known binding to a target
Iterative cycle for improved & new transforms
18
New LHASA
transforms
Implement in
CACTVS
SAVI pilot runs
Manual
inspectio...
19
Future plans
 Stricter criteria or CADD property to qualify a
reaction: less “haystack”
 Develop web-based GUI
 Inve...
CADD Group
• Devendra Dhaked
• Victorien Delannee
CCR
• Nadya Tarasova
Leidos
• Raul Cachau
• Megan Peach
• Jeff Saxe
Else...
Prochain SlideShare
Chargement dans…5
×

Synthetically Accessible Virtual Inventory (SAVI) : Reaction generation and handling at the billion compound scale

133 vues

Publié le

The Synthetically Accessible Virtual Inventory (SAVI) project is an international collaboration between partners in government laboratories, small companies, not-for-profits, and large corporations, to computationally generate a very large number of reliably and inexpensively synthesizable novel screening sample structures. SAVI handles reactions not by virtue of applying simple SMIRKS to a set of building blocks of unknown availability. It instead combines a set of transforms richly annotated with chemical context, coming from, or being newly developed in the mold of, the original LHASA project knowledgebase, with a set of highly annotated, reliably available, purchasable starting materials. These components are tied together for SAVI product generation with the chemoinformatics toolkit CACTVS with custom developments for this project. Each product is annotated with a number of computed properties seen as important in current drug design, including rules for identifying potentially reactive or promiscuous compounds. After having produced and made publicly available the first (beta) set of 283 million SAVI products annotated with proposed one-step syntheses, we will be reporting on the second full production run aimed at creating a database of one billion high-quality, easily synthesizable screening samples. We will present the current status, ongoing developments, as well as scientific and technical challenges of the project.

Publié dans : Santé & Médecine
  • Soyez le premier à commenter

Synthetically Accessible Virtual Inventory (SAVI) : Reaction generation and handling at the billion compound scale

  1. 1. Chemical Biology Laboratory Center for Cancer Research National Cancer Institute National Institutes of Health Synthetically Accessible Virtual Inventory (SAVI) : Reaction generation and handling at the billion compound scale Hitesh Patel, Yuri Pevzner, Wolf D. Ihlenfeldt, Marc Nicklaus
  2. 2. Idea of SAVI 2 What can I make easily, reliably, safely and cheaply? Not in screening databases Many steps and very costly Potentially unsafe Syntheses often unsuccessful $$$$ ∉ Problems of de novo design
  3. 3. Goal of SAVI 3 Reactant A Reactant B SAVI Product you can buy the starting materials 1-step reaction no exotic equipment or dangerous procedures annotate them with computed properties important for CADD
  4. 4. SAVI components 4 CACTVS Lhasa Ltd (UK), LHASA LLC (US) Building Blocks Transforms Chemoinformatics engine LHASA transforms written in CHMTRN/PATRAN
  5. 5. Reaction enumeration tools & libraries based on them  REAL database – Enamine • 145k building blocks X 120 transforms • REAL Space Navigator: BioSolveIT • http://www.enamine.net/index.php?option=com_content&task=view&id=254%3f  CHIPMUNK dataset – TU Dortmund • Hartenfeller1, Reaxys and multicomponent reactions • http://www.ccb.tu-dortmund.de/koch  Scaffold design – ChemPass • https://www.chempassltd.com/synspace-general-scaffold-design  Reactor – ChemAxon • https://chemaxon.com/products/reactor  Other in house tools in industry • Proximal Lilly Collection • Pfizer Global Virtual Library (PGVL) • Boehringer Ingelheim - BI CLAIM 5 1. Hartenfeller M, Eberle M, Meier P, et al (2011) A Collection of Robust Organic Synthesis Reactions for In Silico Molecule Design. J Chem Inf Model 51:3093–3098. doi: 10.1021/ci200379p
  6. 6. CHMTRN/PATRAN transform example 6 DISCONNECTIVE SUBGOAL*ALLOWED BROKEN*BONDS BOND*5 BOND*8 BOND*1 ... IF CARBON ON ALPHA TO ATOM*1 OFFPATH THEN SAVE IT AS 1 KILL IF NOT ARYL ON SAVED*ATOM 1 AND:IF & SAVED*ATOM 1 IS MULTIPLY BONDED ...Possible elimination. ...Would eliminate. KILL IF WITHDRAWING BOND ON ATOM*6 OFFPATH KILL IF WITHDRAWING BOND ON ATOM*8 OFFPATH KILL IF AROMATIC BOND ON ATOM*6 OFFPATH IF BOND*6 IS A FUSION BOND THEN SUBTRACT 10 AND*THEN KILL IF & NOT IN A RING OF SIZE 6 SUBTRACT 10 IF SECONDARY*CENTRE ON SAVED*ATOM 1 SUBTRACT 10 IF TERTIARY*CENTRE ON SAVED*ATOM 1 ...Steric hindrance. IF FEWER THAN TWO HYDROGENS ON ATOM*3 THEN SUBTRACT 15 ...Works best with acetoacetic ester. SUBTRACT 10 IF NOT HYDROGEN ON ATOM*1 ADD 15 IF ARYL ON ALPHA TO ATOM*6 OFFPATH ADD 10 IF ARYL ON ATOM*8 OFFPATH ...Higher yields. IF THERE IS A FUNCTIONAL GROUP ON ATOM*5 THEN DESIGNATE & IT AS PARTICIPATING CONDITIONS SnCl4/25 ACTUAL*CONDITIONS 419: ZnCl2 .... BREAK BOND*5 BREAK BOND*8 SINGLE BOND*6 ATTACH A KETONE ON ATOM*7 ATTACH AN ALCOHOL ON ATOM*6 BOND*1 IS DEFINED*SYN TO BOND*4 ... TRANSFORM 1039 NAME Feist Synthesis of Pyrroles REFERENCES J. Org. Chem. [53], 2084 (1988); D.M. McKinnon, Can. J. Chem. [43], 2628 (1965). COOR _______/ II II II II ===> AR-COCH-C(H) + (H)C-NHC=CHCO2R / / I I (AR) C N C OH CH3 END*REFERENCES ... TYPICAL*YIELD POOR RELIABILITY FAIR REPUTATION FAIR HOMOSELECTIVITY GOOD HETEROSELECTIVITY FAIR ORIENTATIONAL*SELECTIVITY FAIR CONDITION*FLEXIBILITY FAIR THERMODYNAMICS FAIR 1D*PATTERN N%C(-C[FGNOT=LEAVING])%C(- C[FGS=ESTER,KETONE,NITRILE,AMIDE])% C[HETS=0;FGNOT=LEAVING]%C(- C[FGNOT=ACID,LEAVING,XWITHDRAWING])%@1 2D*PATTERN C COR,CHO,CN,COOR,CON H C COR,CHO,COOR,CN,CON H O | | | | | | " N%C%C%C%C%@1 => N-C=C C- -C | | | | C H OH C END*PATTERNS
  7. 7. CHMTRN/PATRAN transforms: smarter than SMARTS 7 24 14 IF BOND*6 IS A FUSION BOND THEN SUBTRACT 10 34 ADD 10 IF ARYL ON ATOM*8 6 7 8 IF BOND*6 IS A FUSION BOND THEN SUBTRACT 10 AND*THEN KILL IF & NOT IN A RING OF SIZE 6 KILL IF WITHDRAWING BOND ON ATOM*6 OFFPATH 49 ADD 15 IF ARYL ON ALPHA TO ATOM*6 ADD 10 IF ARYL ON ATOM*8 24 IF BOND*6 IS A FUSION BOND THEN SUBTRACT 10 ADD 10 IF ARYL ON ATOM*8
  8. 8. Workflow 8 Building blocks Pattern matching building blocks for each transform Transforms reactant1 building blocks reactant2 building blocks Product(s) Are reactants same as input ? Scored lhasa reaction Ignore the reaction Save product, reaction and reactants Filter by mol. wt., possibility of getting mixture Is there a scored lhasa reaction? Skip the product Get retro-synthetically scored reactions Generate product Based on pattern in transform Make reactant pairs No Yes No Yes Reaction killed
  9. 9. First SAVI production run (2016) 9 • CHMTRN/PATRAN transform logic disqualified 17% of reactions • Unique products Downloadable at: https://cactus.nci.nih.gov/download/savi_download/ 377,474 building blocks 14 transforms X351,298,493 305,775,262 283,194,312
  10. 10. 10 Product property distributions 0 5 10 15 20 100 200 300 400 500 600 700 800 900 1000 >1000 Millions Mol Wt. Molecular weight 0 1 2 3 4 5 0 100 200 300 400 500 600 700 800 900 1000 1100 1200 1300 1400 1500 1600 1700 1800 1900 2000 >2000 Millions USD/gm Cost per gram (predicted based on unit price) 0 2 4 6 8 10 12 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 Millions Fraction of SP3 Carbons Fraction of SP3 hybridized carbons
  11. 11. SAVI produces novel ring systems Total novel SA rings/scaffolds created: 999 (1) Ertl et al., J. Med. Chem. 49, 4568–4573 (2006) “Novel”: not found in CADD Group aggregated database of 152 million structures Simple Aromatic (SA) “Ertl type”(1) novel ring/scaffold systems SA: Systems up to tricyclic, ≥ 1 aromatic ring, with hetero atoms O, N, S containing the moieties: Novelty confirmed in SciFinder, few examples: 11
  12. 12. 12 Annotations for drug design  “Drug design” properties: • RO5 violations, PAINS matches, fsp3, log P… • All 275 Bruns & Watson rules1 - combined as a demerit score  Sigma catalog numbers and unit price of building blocks  Proposed reaction > <SAVI_BUILDING_BLOCK_A_ORDER_LINK> http://www.aldrichmarketselect.com/listBrowser.asp?structure_id=65139341 > <SAVI_BUILDING_BLOCK_A_COST_GRAM> 482.0 1. Bruns RF, Watson IA (2012) Rules for identifying potentially reactive or promiscuous compounds. J Med Chem 55:9763– 9772. doi: 10.1021/jm301008n > <SAVI_REACTION> Paal-Knorr Pyrrole Synthesis > <SAVI_LHASA_SCORE> 45 > <SAVI_PREDICTED_YIELD> VERY*GOOD
  13. 13.  ~600k building blocks  ~20 chemistries (~50 transforms) implemented • 7 entirely new chemistries • example of new chemistry: Suzuki-Miyaura coupling 13 Next version of SAVI OH / Ar' - [Cl,Br,I] + Ar - B ====> Ar - Ar' OH Ar - I + HC#R ====> Ar - C#R' Br R1 OH Ar R1 / / / C=C + Ar-B ====> C=C R2 OH R2 Suzuki coupling of Chloro, Bromo, Iodo Suzuki coupling of Alkenes Suzuki coupling with Alkynes
  14. 14. 14 1 M 10 M 100 M 1 B 10 B 100 B 100 K 10 K 1 K ~100 Billion total possible pairs Possible reactant pairs: >> 1 billion
  15. 15. 15  To generate comparable number of products for each transform  Select 1 billion truly “interesting” compounds from 100 billions reactions  What filter criteria? Possible: • Molecular weight, QED score1 of products • Availability, reliability, purity and price of BBs • Strict scoring by SUBTRACT -> KILL  How to handle possible mixtures of products  Handling and representation of 1 billion reactions Poster (P 62) - The need for comprehensive reaction handling in SAVI and beyond, Marc Nicklaus Challenges 1. Bickerton GR, Paolini G V, Besnard J, et al (2012) Quantifying the chemical beauty of drugs. Nat Chem 4:90–98. doi: 10.1038/nchem.1243
  16. 16. More and better annotations planned 16 Predicted reaction conditions Determine if SAVI product is a mixture Annotate with possible targets 2nd step [Na]N=[N+]=[N-]
  17. 17. Ongoing work 17 grow molecule elsewhere Preserve reactant’s functional groups already known binding to a target
  18. 18. Iterative cycle for improved & new transforms 18 New LHASA transforms Implement in CACTVS SAVI pilot runs Manual inspection by synthetic chemists and CADD group Prioritizing updates and improvements
  19. 19. 19 Future plans  Stricter criteria or CADD property to qualify a reaction: less “haystack”  Develop web-based GUI  Investigate multi-step reactions  Combine with robotic chemistry
  20. 20. CADD Group • Devendra Dhaked • Victorien Delannee CCR • Nadya Tarasova Leidos • Raul Cachau • Megan Peach • Jeff Saxe Elsevier • Scott Hutton • Matt Clark InfoChem • Hans Kraut • Josef Eiblmaier Merck KGaA • Friedrich Rippmann • Mireille Krier • Tim Knehans MilliporeSigma • Bret Daniel • Chad Hurwitz • Subir Ghorai • Daniel Boesch • Shashi Jasty Lhasa Limited, UK • Philip Judson • Martin Ott LHASA LLC, U.S. • Alan Long • Alex Sukharevsky Novartis • Peter Ertl National Institute of Chemistry, Slovenia • Dušanka Janežič • Janez Konc 20 Acknowledgements

×