Prediction is much harder than analysis. Consider hurricanes and tornadoes; it's much easier to follow the path of destruction by locating devastated neighborhoods, than to forecast the paths of such weather systems in advance. Likewise for many chemical reactions, such as nitration (by refluxing with nitric acid and sulfuric acid) where the appearance of one or more nitro groups indicates a nitration reaction, but predicting where on a non-trivial organic molecule this functional group appears is a much harder challenge. In this sense, reaction analysis is much simpler than (either forward or retrosynthetic) synthesis planning.
NextMove Software's namerxn is an expert system for classifying reactions (from reaction SMILES, MDL connection tables or ChemDraw sketches) typically assigning each reaction instance to a leaf classification in the Royal Society of Chemistry's RXNO ontology. These tools can be helpful in the analysis of regioselectivity preferences of reactions.
This talk consists of two parts. A technical part describing the recent algorithmic and methodological improvements to the namerxn software, including describing some of the more challenging of the 1000+ reactions it currently identifies. And a scientific part that investigates the regioselective preferences of some of these reactions.
Module for Grade 9 for Asynchronous/Distance learning
CINF 170: Regioselectivity: An application of expert systems and ontologies to chemical (named) reaction analysis
1. Regioselectivity: An Application of Expert
Systems and Ontologies to Chemical
(Named) Reaction Analysis
CINF Reaction Analytics. 256th ACS National Meeting, Boston, MA. Thursday 23rd August 2018
Roger Sayle, John Mayfield and Noel O’Boyle
NextMove Software, Cambridge, UK
2. analysis vs. prediction
• In “Applied Chemoinformatics” (2018), J. Goodman
defines three main problems.
– Reaction Planning: R → ? → P [Database]
– Reaction Prediction: R1 + R2 → ? [Simulation]
– Synthesis Planning: R? + R? + R? → P [Design]
• A corollary is that there’s a distinction between
reactions that have already been observed and
those experiments yet to be performed.
CINF Reaction Analytics. 256th ACS National Meeting, Boston, MA. Thursday 23rd August 2018
3. analysis vs. prediction
CINF Reaction Analytics. 256th ACS National Meeting, Boston, MA. Thursday 23rd August 2018
4. intuition and counter-intuition
CINF Reaction Analytics. 256th ACS National Meeting, Boston, MA. Thursday 23rd August 2018
With apologies to Mark Twain:
“Never let the facts get in the
way of a good synthesis plan”.
There are reactions that chemists
expect will happen but don’t &
those they don’t expect but do.
But cheminformaticians are
alchemists that can turn lead into
gold, as easily as “[Pb]>>[Au]”.
5. Experimental validation
• Synthesis of a novel aromatic
heterocycle previously unreported in
the literature.
• William Pitt et al., “Heteroaromatic
Rings of the Future”, Journal of
Medicinal Chemistry, 52(9):2952-
2963, 2009.
CINF Reaction Analytics. 256th ACS National Meeting, Boston, MA. Thursday 23rd August 2018
6. expectations: setting the scope
Goodman Challenges Carey et al. Challenges
Maitotoxin
Difficult to access substituted
aromatic starting materials.
Eribulin Org. Biomol. Chem. 2005, 4,2337-2347.
CINF Reaction Analytics. 256th ACS National Meeting, Boston, MA. Thursday 23rd August 2018
7. Which reactions are important?
• NCI/LHASA SAVI (14+6 reactions)
– Suzuki Coupling 36262 out of 1.2M USPTO examples
– Sulfonamide Schotten-Baumann 14348 out of 1.2M USPTO examples
– Buchwald-Harwig 6040 out of 1.2M USPTO examples
– Hiyama coupling 458 out of 1.2M USPTO examples
– Fukuyama coupling 2 out of 1.2M USPTO examples
– Liebeskind-Srogl coupling 0 out of 1.2M USPTO examples
• Hartenfeller/Schneider (58 reactions)
– #1 Pictet-Spengler reaction 7 out of 1.2M USPTO examples
– #10+ #11 Azide-nitrile Huisgen-cycloaddition 5 out of 1.2M USPTO examples
– #17 Pyridone synthesis 2 out of 1.2M USPTO examples
– #20 Phthalazinone synthesis 16 out of 1.2M USPTO examples
– #24 Friedlander quinoline synthesis 30 out of 1.2M USPTO examples
• Enamine REAL 2016 (43 reactions)
– Thiourea to guanidine (14 out of 160M examples).
CINF Reaction Analytics. 256th ACS National Meeting, Boston, MA. Thursday 23rd August 2018
8. myth #1: heterocycle formation
• Because almost all drug-like molecules contain a
heterocycle, there’s a belief that heterocycle
formation is important.
• Analysis of both patent data and pharmaceutical
ELNs reveals that heterocycle forming reactions,
even named heterocycle forming reactions, are
relatively rare, with ring systems often being
purchased as building blocks*.
* This analysis might not apply to process development and manufacturing.
https://nextmovesoftware.com/blog/2016/10/24/buying-a-ring-or-making-one-yourself/
CINF Reaction Analytics. 256th ACS National Meeting, Boston, MA. Thursday 23rd August 2018
9. Reactions that don’t happen!
• Paal-Knorr Pyrrole Synthesis vs. Aldehydes/Ketones
Of the 430 examples of
Paal-Knorr pyrrole
synthesis reported in
US patent applications
2001-2012, exactly
zero have more than
the two reacting
ketones/aldehydes.
CINF Reaction Analytics. 256th ACS National Meeting, Boston, MA. Thursday 23rd August 2018
10. Who let the dogs out?
• Big Data can determine the utility and scope of reactions.
CINF Reaction Analytics. 256th ACS National Meeting, Boston, MA. Thursday 23rd August 2018
11. Reactions that do happen!
• Chloro Sonogashira Couplings (@ NCI)
– The LHASA ‘CHMTRN’ rules for Transform 2267, Sonogashira couplings,
(presented by Marc Nicklaus at Sheffield) states “Iodides are usually
more reactive than bromide. Chlorides do not react”.
Code Name Count Yield
3.3.2 Bromo Sonogashira coupling 3717 49.3%
3.3.3 Chloro Sonogashira coupling 429 44.2%
3.3.4 Iodo Sonogashira coupling 2721 64.9%
• Isotopically-labelled Compounds (@Eli Lilly)
– The LAAR reactions used to construct Lilly’s PLC (Nicolaou et al. 2016)
forbid the presence of isotopic labels in reactants.
CINF Reaction Analytics. 256th ACS National Meeting, Boston, MA. Thursday 23rd August 2018
12. improving upon synthia™
• As presented earlier, Chematica/Synthia maintains a
manually curated “black list” of interfering functional
groups.
• A more labor-efficient data-driven strategy is to
automatically maintain a “white list” of tolerated
functional groups that can be derived from observed
experiments.
• It’s much easier to track the things you know about
and can see, than the things you can’t see and/or
don’t know about.
CINF Reaction Analytics. 256th ACS National Meeting, Boston, MA. Thursday 23rd August 2018
13. handling noisy data (n>1 statistics)
• Kumada couplings incompatible with aldehydes...
• US20050107257A1 [p0196] Syngenta
A solution of 34.3 ml of ethylmagnesium chloride in 50 ml of tetrahydrofuran is added dropwise at −70° C. to
7.3 g of indium trichloride in 200 ml of tetrahydrofuran and, after stirring for 30 minutes, the reaction
temperature is allowed to rise slowly to room temperature. That solution is added to a solution of 14.9 g of 3-
bromo-4-pyridine carbaldehyde and 2.8 g of PdCl2 (PPh3)2 in 240 ml of tetrahydrofuran and the reaction
mixture is heated under reflux for 20 hours. 5 ml of methanol are then added and the mixture is concentrated
in vacuo, stirred thoroughly with diethyl ether, filtered off and concentrated in vacuo once more. The residue is
chromatographed on silica gel using ethyl acetate/hexane (1:1), yielding 3-ethyl-4-pyridine carbaldehyde (I) in
the form of a yellow oil.
CINF Reaction Analytics. 256th ACS National Meeting, Boston, MA. Thursday 23rd August 2018
14. c.f. US20140296184A1 [C00008]
M. Segler, M. Preuss and M. Waller, “Planning Chemical Syntheses with Deep Neural
Networks and Symbolic AI”, Nature, 555:604-610, 2018.
A.Gini, M. Segler et al, “Dehydrogenative TEMPO-mediated formation of Unstable
Nitrones: Easy Access to N-Carbamoyl Isoxazolines”, Chem. Eur. J. 21:12053-12060,
2015.
prediction? about the future?
CINF Reaction Analytics. 256th ACS National Meeting, Boston, MA. Thursday 23rd August 2018
15. search: monte carlo v. proof num
• Appropriate choice of AI search technique.
Monte Carlo search is inappropriate
for search problems such as mazes.
White to win in 173 ply
Kf1-f2!
CINF Reaction Analytics. 256th ACS National Meeting, Boston, MA. Thursday 23rd August 2018
OR
AND
0 1 0 0 1 1 1 0
0
1
0 1 0
16. categorization of reactions
1. J. Carey, D. Laffan, C. Thomson, M. Williams, Org. Biomol. Chem. 2337, 2006.
2. S. Roughley and A. Jordan, J. Med. Chem. 54:3451-3479, 2011.
34%
17%
5%
2%
3%
6%
10%
1%
15%
2%
5% Heteroatom alkylation and arylation
Acylation and related processes
C-C bond formations
Heterocycle formation
Protections
Deprotections
Reductions
Oxidations
Functional group conversion
Functional group addition
Resolution
CINF Reaction Analytics. 256th ACS National Meeting, Boston, MA. Thursday 23rd August 2018
18. reaction ontology
• Reactions are classified into a common subset of the
Carey et al. classes and the RSC’s RXNO ontology.
• There are 12 super-classes
– e.g. 3 C-C bond formation (RXNO:0000002).
• These contain 84 class/categories.
– e.g. 3.5 Pd-catalyzed C-C bond formation (RXNO:0000316)
• These contain ~1050 named reactions/types.
– e.g. 3.5.3 Negishi coupling (RXNO:0000088)
• These require ~2200 SMIRKS-like transformations.
CINF Reaction Analytics. 256th ACS National Meeting, Boston, MA. Thursday 23rd August 2018
19. Complication: agents matter
From http://en.wikipedia.org/wiki/Diazonium_compound
Sandmeyer reaction
Benzenediazonium chloride heated with cuprous chloride
disolved in HCl to yield chlorobenzene.
C6H5N2
+ + CuCl → C6H5Cl + N2 + Cu+
Gatterman reaction
Benzenediazonium chloride is warmed with copper powder and
HCl to yield chlorobenzene.
C6H5N2
+ + CuCl → C6H5Cl + N2 + Cu+
CINF Reaction Analytics. 256th ACS National Meeting, Boston, MA. Thursday 23rd August 2018
20. 10 most popular reactions
ID Name Count
2.1.2 Carboxylic acid + amine 26,040
1.3.1 Buchwald-Hartwig amination 22,048
3.1 Suzuki coupling 16,508
1.7.6 Williamson ether synthesis 15,665
2.1.1 Amide Schotten-Baumann 11,016
7.1 Nitro to amino 10,234
6.1.1 N-Boc deprotection 9,821
6.2.2 CO2H-Me deprotection 9,487
6.2.1 CO2H-Et deprotection 6,749
2.2.3 Sulfonamide Schotten-Baumann 6,223
CINF Reaction Analytics. 256th ACS National Meeting, Boston, MA. Thursday 23rd August 2018
21. Most/least successful reactions
ID Name Mean Yield Count
1.7.2 Diazomethane esterification 91% 41
9.3.1 Carboxylic acid to acid chloride 88% 704
9.7.14 Bromo to azido 85% 235
1.7.5 Methyl esterification 84% 2918
9.7.19 Bromo to iodo Finkelstein reaction 82% 116
6.1.3 N-Cbz deprotection 81% 1359
…
4.1.11 Larock indole synthesis 47% 55
3.11.3 Ullmann-type biaryl coupling 44% 407
1.7.1 Chan-Lam ether coupling 44% 154
4.1.4 Pinner pyrimidine synthesis 39% 47
CINF Reaction Analytics. 256th ACS National Meeting, Boston, MA. Thursday 23rd August 2018
24. Namerxn reaction naming
• Many reaction classification algorithms are
dependent upon atom-atom mapping assignments.
• Alas MCS-based atom mapping algorithms are often
slow and/or inaccurate [Lowe & Sayle 2012 & 2013].
• NameRXN is a mechanism-based atom-mapper.
• All reactants and reagents are placed in a single pot
(molecule) and sets of SMIRKS applied in turn.
• If the desired product is generated, the reaction (its
mechanism and mapping) is identified.
– Rationalization is easier than prediction!
CINF Reaction Analytics. 256th ACS National Meeting, Boston, MA. Thursday 23rd August 2018
28. Recent improvements insights
• Profiling NameRxn in 2014 to diagnose performance
problems with RDKit revealed that pattern matching
and transformation accounted for <1% of runtime.
• The bottleneck is actually in canonicalization.
• The Ah-ha experience was to use hash filtering.
• Check molecular formula: CiHjBrkCllFmInNoOpPqSr
• Additional cleverness allows pre-sanitization hashing.
– Triple bond count, but not single or double bond count.
– Perhaps there’s something in InChI-style hashing after all.
CINF Reaction Analytics. 256th ACS National Meeting, Boston, MA. Thursday 23rd August 2018
34. tactical vs. strategic reactions
• Traditional Synthesis
Planning has concentrated
on the Strategic Application
of Named Reactions.
• However, there’s much to
be gained for the Tactical
Application of Unnamed
Reactions.
• Nitro reduction is the 6th
most frequent reaction.
CINF Reaction Analytics. 256th ACS National Meeting, Boston, MA. Thursday 23rd August 2018
35. functional group interconversion
• Relative frequency of simple group conversion
From a total of 7,252,419 reactions from USPTO & EPO patents.
CINF Reaction Analytics. 256th ACS National Meeting, Boston, MA. Thursday 23rd August 2018
Amino Bromo Chloro Fluoro Hydroxy Iodo Sulfanyl Thioxo
Amino 3951 2949 990 3976 3657 143
Bromo 3121 589 637 2390 738 435
Chloro 9424 717 1606 2156 744 798
Fluoro 1549 180 484 826 28 103
Hydroxy 2572 11441 31593 7641 3004 348
Iodo 155 445 47
Nitro 126606 138
Oxo 8419
36. nextmove’s strategy
• Why expose reaction planning to the end-user at all?
• During our collaboration with ChemSpace, it has become clear
that what was required was not similarity nor superstructure
search by a form of synthesis-aware search.
• This is similar to the challenge faced by traditional
restrosynthesis tools in identifying a leaf/goal state.
• 937M purchasable compounds makes this is non-trivial.
• The usual challenges of functional group, tautomer and
protonation state lookups, now also protecting groups.
• But why not return the carboxylic acid when searching for the
acid chloride, or alcohol when searching for bromo derivative.
CINF Reaction Analytics. 256th ACS National Meeting, Boston, MA. Thursday 23rd August 2018
37. building block search
• Query:
Results:
CINF Reaction Analytics. 256th ACS National Meeting, Boston, MA. Thursday 23rd August 2018
38. the challenge of regioselectivity
• A tricky benchmark is reactions of 2,4,5-trichloropyrimidine
• The nature of pyrimidine makes the chloro at the 4-position more reactive
than the 2 position which is more reactive than the 5 position.
• Simple quantum mechanical have difficulty discerning this order.
CINF Reaction Analytics. 256th ACS National Meeting, Boston, MA. Thursday 23rd August 2018
39. handy’s rule
• Scott Handy and Yanan Zhang, “A Simple Guide for Predicting
Regioselectivity in the Coupling of Polyhaloheteroaromatics”,
Chemical Communications, 3:299-301, Nov 2005.
Abstract
A simple guide for predicting the order and site of coupling (Suzuki, Stille, Negishi,
Sonogashira, etc.) in polyhaloheteroaromatics based upon the 1H NMR chemical shift
values of the parent non-halogenated heteroaromatics has been developed.
CINF Reaction Analytics. 256th ACS National Meeting, Boston, MA. Thursday 23rd August 2018
40. electrophilic substitution of
heterocycles with qm methods
• J.C. Kromann et al., “Fast and Accurate Prediction of the
Regioselectivity of Electrophilic Aromatic Substitution
Reactions”, Chem. Sci., 9(3):660-665, Nov 2017.
• M. Kruszyk et al., “Computational Methods to Predict the
Regioselectivity of Electrophilic Aromatic Substitution
Reactions of Heteroaromatic Systems”, J. Org. Chem.
81(12):5128-5134, Jun 2016.
CINF Reaction Analytics. 256th ACS National Meeting, Boston, MA. Thursday 23rd August 2018
41. a data-driven strategy
• One possible approach to the challenge of regioselectivity is
to derive preferences by large-scale (statistical) analysis of
reaction data sets.
• Reaction classification can identify the subset of relevant
examples, which can then be used to produce tables of
heterocycles position preferences.
• Directing groups and their influence can also be identified and
tabulated.
• See https://www.scripps.edu/baran/images/grpmtgpdf/Gutekunst_Apr_10.pdf
CINF Reaction Analytics. 256th ACS National Meeting, Boston, MA. Thursday 23rd August 2018
42. the semantics of atom mapping
Prof. Goodman poses an interesting question of atom mapping.
4.1.6 Cyclic Beckman Rearrangement
1.2.9 Alcohol + Amine Condensation or 1.1.3 Iodo N-methylation
CINF Reaction Analytics. 256th ACS National Meeting, Boston, MA. Thursday 23rd August 2018
43. Acknowledgements
• NextMove Software
– Noel O’Boyle
– John Mayfield
• NextMove Alumni
– Daniel Lowe
• Thank you for you time.
• Questions?
• Thoughts?
• AbbVie
• AstraZeneca
• Bristol-Myers Squibb
• ChemSpace
• Eli Lilly
• GlaxoSmithKline
• Hoffmann-La Roche
• IBM Research Zurich
• Merck
• Novartis
• Royal Society of Chemistry
• Vernalis
• Vertex Pharmaceuticals
CINF Reaction Analytics. 256th ACS National Meeting, Boston, MA. Thursday 23rd August 2018
44. Transforms vs. reactions
Importance of Reaction Mechanism
Example: Ullman-type Coupling Reactions
SMIRKS: [H][N:1].[Cl][c:2]>>[N:1][c:2]
The SMIRKS transform alone is insufficient
to predict the products and by-products in
this example. A measure of nucleophility
is desirable for each atom in a molecule.
Without this software may be misled into
believing that protecting groups are required.
CINF Reaction Analytics. 256th ACS National Meeting, Boston, MA. Thursday 23rd August 2018
45. Analysis of pharmaceutical elns
• NextMove Software’s HazELNut software is used to
export and analyze ELN content at 6 of the top 10
large pharmaceutical companies.
• In-house analysis of this data, across the industry,
reveals a surprisingly high rate of synthesis failure,
not indicated in the published literature (journal
articles, patent applications or reaction databases).
• Understanding the causes of these failures is perhaps
more significant than attempting to access new
chemistries.
CINF Reaction Analytics. 256th ACS National Meeting, Boston, MA. Thursday 23rd August 2018
46. Extracting mps and reactions
CINF Reaction Analytics. 256th ACS National Meeting, Boston, MA. Thursday 23rd August 2018
47. Example reaction mining INPUT
Methyl 4-[(pentafluorophenoxy)sulfonyl]benzoate
To a solution of methyl 4-(chlorosulfonyl)benzoate (606 mg,
2.1 mmol, 1 eq) in DCM (35 ml) was added
pentafluorophenol (412 mg, 2.2 mmol, 1.1 eq) and Et3N
(540 mg, 5.4 mmol, 2.5 eq) and the reaction mixture stirred
at room temperature until all of the starting material was
consumed. The solvent was evaporated in vacuo and the
residue redissolved in ethyl acetate (10 ml), washed with
water (10 ml), saturated sodium hydrogen carbonate (10
ml), dried over sodium sulphate, filtered and evaporated to
yield the title compound as a white solid (690 mg, 1.8
mmol, 85%).
49. CHEMICAL reactions for free
CINF Reaction Analytics. 256th ACS National Meeting, Boston, MA. Thursday 23rd August 2018
50. bond changes in indole synthesis
Synthesis A B C D
Baeyer-Emmerling M
Bartoli M M
Bischler-Möhlau M C M
Fischer M C M
Fukuyama M
Hemetsberger M
Larock M C M
Mandelung M
Nenitzescu M M
Reissert M M
CINF Reaction Analytics. 256th ACS National Meeting, Boston, MA. Thursday 23rd August 2018
51. Suzuki coupling leaving groups
Leaving Group Mean Yield N Observations
Bromo 58.80% 10817
Chloro 57.96% 2752
Iodo 57.21% 2049
Triflyloxy 65.48% 717
CINF Reaction Analytics. 256th ACS National Meeting, Boston, MA. Thursday 23rd August 2018
52. Beyond drug guru
sildenafil (viagra) vardenafil (levitra)
CINF Reaction Analytics. 256th ACS National Meeting, Boston, MA. Thursday 23rd August 2018
53. Eli lilly’s automated synthesis lab
Alexander G. Godfrey, Thierry Masquelin and Horst Hemmerle, “A Remote-Control Adaptive MedChem Lab: An Innovative
Approach to Enable Drug Discovery in the 21st Century”, Drug Discovery Today, Vol. 18, Nos. 17-18, September 2013.
CINF Reaction Analytics. 256th ACS National Meeting, Boston, MA. Thursday 23rd August 2018
54. Synthesis failures at lilly
• At the 2013 Sheffield Cheminformatics conference,
Christos Nicolaou highlighted the technical challenge
with predicting compounds potentially accessible by
the Lilly’s Advanced Synthesis Lab (ASL).
• In a proof-of-concept pilot project, only 25 of 90
compounds suggested by Lilly’s Annotated Reaction
Repository (LARR) rule-set could be successfully
synthesized in practice.
• http://cisrg.shef.ac.uk/shef2013/talks/14.pdf
CINF Reaction Analytics. 256th ACS National Meeting, Boston, MA. Thursday 23rd August 2018
55. Synthesis failures at gsk
• Pickett et al. 2011 describe the parallel synthesis of a
50x50 library of MMP-12 inhibitors by an iodo-Suzuki
coupling reaction.
• Only 1704 of 2500 could be assayed [566 not made]
Pickett et al., ACS Med. Chem. Lett. 2(1):28, 2011
CINF Reaction Analytics. 256th ACS National Meeting, Boston, MA. Thursday 23rd August 2018
56. Learning from failure (logp shift)
• Nadin et al. 2012 [1] hypothesize that low LogP is a
major cause of synthesis failure in parallel synthesis
of combinatorial libraries.
• Analysis confirms that this is indeed a significant
factor for the GSK MMP-12 library.
– 1704 compounds measured, mean logP = 3.56 (1.44)
– 566 compounds not made, mean logP = 2.83 (1.52)
– Student’s t-test for different distributions, p<2x10-22.
1. Nadin, Hattotuwagama and Churcher ,“Lead-Oriented Synthesis: A New Opportunity for
Synthetic Chemistry”, Angew. Chem. Int. Ed, 51:1114 2012.
CINF Reaction Analytics. 256th ACS National Meeting, Boston, MA. Thursday 23rd August 2018
57. Example (actionable) insight
The clear trend between Suzuki coupling success rate and
predicted octanol-water partition co-efficient.
Data: Pickett et al., ACS Med. Chem. Lett. 2(1):28, 2011
97 232
340
525
474 202
63
55 119 127 141 80 36 8
0%
10%
20%
30%
40%
50%
60%
70%
80%
90%
100%
<1.0 1.0-2.0 2.0-3.0 3.0-4.0 4.0-5.0 5.0-6.0 >6.0
Reaction Product Predicted cLogP
Sucessful Reactions Failed Reactions
CINF Reaction Analytics. 256th ACS National Meeting, Boston, MA. Thursday 23rd August 2018
58. big data mining confirmation of
Nadin-churcher hypothesis
On 16,335 Suzuki coupling reactions extracted from US
patent applications between 2001 and 2012.
Nadin, Hattotuwagama and Churcher ,“Lead-Oriented Synthesis: A New Opportunity for Synthetic
Chemistry”, Angew. Chem. Int. Ed, 51:1114 2012.
LogP Mean Yield N Obs
< 1.0 52.89% 196
1.0 – 2.0 56.02% 1155
2.0 – 3.0 56.72% 2881
3.0 – 4.0 58.14% 4071
4.0 – 5.0 57.26% 3186
5.0 – 6.0 59.25% 2126
> 6.0 63.83% 2720
CINF Reaction Analytics. 256th ACS National Meeting, Boston, MA. Thursday 23rd August 2018
59. “big data” reaction yield analysis
AZ Data courtesy of Nick Tomkinson, AstraZeneca RDI, Alderley Park, UK.
60. “big data” reaction yield analysis
AZ Data courtesy of Nick Tomkinson, AstraZeneca RDI, Alderley Park, UK.