AiZynth Impact on Medicinal Chemistry Practice at AstraZeneca

Jason D. Shields, Rachel Howells, Gillian Lamont, Yin Leilei, Andrew Madin, Christopher E. Reimann, Hadi Rezaei, Tristan Reuillon, Bryony Smith, Clare Thomson, Yuting Zhengc and Robert E. Ziegler (2024)
Highlighted by Jan Jensen

Figure 3 from this paper (c) the authors 2020. Reproduced under the CC-BY license

This is one of the rare papers where experimental chemists talk candidly about their experiences using ML models developed by others. In this case it is AiZynthFinder, which is developed at AstraZeneca Gothenburg and predicts retrosynthetic paths, while the users are most synthetic chemists at AstraZeneca in the UK, US, and China. The paper is really well written and well worth reading. I'll just include a few quotes below to whet your appetite.  

"New users of AI tools in general are often disappointed by the failure of AI to live up to their expectations, and chemists' interaction with AiZynth is no exception. The first molecule that most new users test is one that they have personally synthesised recently, and AiZynthFinder rarely replicates their route exactly. Due in part to our self-imposed requirement to run fast searches, AiZynthFinder often gets close to a good route. Thus, experienced users seek inspiration from AiZynth rather than perfection."

"Common problems include proposals that would lead to undesired regioselectivity, functional group incompatibility, or overgeneralisation of precedented reactions to an inappropriate context."

"Early problems also included protection/deprotection cycles, which had to be intentionally penalised in order to focus AiZynth on productive chemistry. We have found that protecting group strategy is still best decided by the chemist. Thus, the AI proposals discussed in the case studies do not make heavy use of protecting groups, whereas several of the laboratory syntheses do."

This work is licensed under a Creative Commons Attribution 4.0 International License.

Growing strings in a chemical reaction space for searching retrosynthesis pathways

Federico Zipoli, Carlo Baldassari, Matteo Manica, Jannis Born, and Teodoro Laino (2023)
Highlighted by Jan Jensen

Part of Figure 10 from the paper. (c) The authors 2023. Reproduced under the CC-NC-ND

Prediction of retrosynthetic reaction trees are typically done by stringing together individual retrosynthetic steps that have the highest predicted confidences. The confidence is typically related to the frequency of the reaction in the training set. This approach has two main problems that this paper addresses. One problem is that "rare" reactions are seldom selected even if they might actually be the most appropriate for a particular problem. The other problem is that you only use local information and "strategical decisions typical of a multi-step synthesis conceived by a human expert".

This paper tries to address these problems by doing the selection of steps differently. The key is to convert the reaction (which are encoded as reaction SMILES) to a fingerprint, i.e. a numerical representation of the reaction SMILES, and using them to compute similarity scores.

For example, in the first step you can use the fingerprint to ensure a diverse selection of reactions to start the synthesis of. In subsequent steps, you can concatenate the individual reaction fingerprints (i.e. the growing string) to compute similarities to reaction paths, rather than individual steps. By selecting paths that are most similar to the training data you could incorporate the "strategical decisions typical of a multi-step synthesis conceived by a human expert". Very clever!

The main problem is how to show that this approach produces better retrosynthetic predictions. Once metric might be shorter paths and the authors to note this but I didn't see any data and it's not necessarily the best metric since, for example important protection/deprotection steps could be missing. The best approach is for synthetic experts to weigh in, but that's hard to do for enough reactions to get good statistics. Perhaps this recent approach would work?

This work is licensed under a Creative Commons Attribution 4.0 International License.

The Synthesizability of Molecules Proposed by Generative Models

Wenhao Gao and Connor W. Coley (2020)

Highlighted by Jan Jensen

Figure 1 from the paper. (c) The authors 2020. The paper tests method c, d, and e

Disclaimer: I implemented one of the methods (graph based GA) being tested. 

It is well known that generative models (including genetic algorithms) can suggest very weird-looking molecules when used to optimise molecular properties. This is the first paper that I have come across that tries to quantify this problem by computing their synthesizability.

A molecule is defined as synthesizable if a computer-assisted synthesis planning (CASP) program can find a synthetic route to the molecule. The CASP program they used (ASKCOS) can find synthetic routes for between 57-89% of molecules sampled from commonly used databases (or subsets) such as ChEMBL and ZINC. These databases generally contain molecules that have been made, so just because ASKCOS can't figure out how to make it doesn't mean it can't be made.

The authors used ASKCOS to determine the fraction of synthesizable molecules suggested by three generative models (one ML-based and two GA-based methods) for several "hard" optimisation problems. The ML-based method tends to predict higher fractions of synthesizable molecules compared to GAs and for some properties none of the 100 top-scoring molecules suggested by the GAs were deemed synthesizable. 

The authors go on to show that, in many cases,  the fraction of synthesizable molecules can be increased significantly by including an empirical synthesizability measure in the scoring function, which is very welcome news to me. Furthermore, the top synthesizable molecules shown in the paper look very reasonable, which suggests that CASP programs can weed out the crazy structures.

One worry is that CASP programs are overly conservative and weed out viable structures that could teach us some genuinely new chemistry, but if generative models are to be taken seriously we obviously need a method to exclude the crazy molecules before we show them to synthetic chemists.

This work is licensed under a Creative Commons Attribution 4.0 International License.

Computationally Augmented Retrosynthesis: Total Synthesis of Paspaline A and Emindole PB

Daria E. Kim, Joshua E. Zweig and Timothy R. Newhouse (2018)

Highlighted by Jan Jensen

Figure 2 from the paper reproduced under the CC-BY-NC-ND licence

This paper presents a rare example of using quantum chemical TS calculations to guide, rather than post-rationalise, organic synthesis. The authors wanted to design a retrosynthetic path that could be used to make two related natural products, paspaline A and emindole PB, that require either a ring closure (paspaline A) or a methyl shift (emindole PB). Three different routes were possible that lead to different functionalities that were relatively distant from the ring closure/methyl shift, which made it hard to predict the best route by chemical intuition.

Instead the authors used mPW1PW91/6-31+G(d,p)//B3LYP/6-31G(d) to find the TSs for both reactions for each of the three routes to predict the best route, which turns out to be "C". Route C did indeed work great in practice, while route A (predicted to be worst route) didn't give the desired results.

My guess is that the key here is that the synthetic question was reduced to a question of relative barrier heights of closely related reactions, i.e. ΔΔΔG^‡ = ΔΔG^‡(4→5) - ΔΔG^‡(4→6), which leads to maximum error cancellation. I hope this paper will lead to more use of QM to guide synthetic decisions and more work on making TS calculations even more accessible to synthetic chemists


This work is licensed under a Creative Commons Attribution 4.0 International License.

Planning chemical syntheses with deep neural networks and symbolic AI

Marwin H. S. Segler, Mike Preuss, Mark P. Waller (2018)
Highlighted by Jan Jensen

Figure 1 from the paper. Copyright 2018 Springer Nature

The paper uses a Monte Carlo tree search (MCTS) algorithm (also used in AlphaGo Zero) to suggest retrosynthetic routes that were just as good as those proposed by expert organic chemist. Remarkably the underlying "expert knowledge" is automatically extracted from reaction databases into three neural networks. Thus, the method is referred to as 3N-MCTS.

At the core of this approach are two neural networks that can predict the probability of a molecule undergoing one of either 301,671 or 17,134 chemical transformations, the latter being more computationally efficient than the former. The networks were trained on tranformation rules from 12.4 million single-step reactions from the Reaxys chemistry database, i.e. determined automatically without human intervention.

  

The retrosynthetic "game" is won if the target molecule can be completely decomposed into predefined precursor molecules within 25 retrosynthetic steps, where the 50 most probable chemical transformations are considered for each step. It is not practically possible to test all $50^{25} \approx 10^{40}$ possible retrosynthetic paths so a MCTS is used to search for the best path.

A MCTS starts by evaluating a number of paths randomly and then assigning likelihood scores to the early parts of the paths depending on whether the paths lead to winners or not. The process is then repeated except that the early steps in the path are chosen based on likelihood scores, which are continuously updated and added to unscored steps.  The changing likelihood scores means that the search for new paths is directed towards the more promising areas of the path tree. I have given a short illustration of the process here. The process is repeated for a given number of steps and the path with the best set of likelihood scores is selected.

One of the tests of the method was a double blind study where experienced synthetic chemists were asked to choose between retrosynthetic routes developed by experts and by 3N-MCTS. The study found no clear preference!

I couldn't find any information about code availability.

This work is licensed under a Creative Commons Attribution 4.0 International License.

Equatorenes: Synthesis and Properties of Chiral Naphthalene, Phenanthrene, Chrysene, and Pyrene Possessing Bis(1-adamantyl) Groups at the Peri-position

Yamamoto, K.; Oyamada, N.; Xia, S.; Kobayashi, Y.; Yamaguchi, M.; Maeda, H.; Nishihara, H.; Uchimaru, T.; Kwon, E. J. Am. Chem. Soc. 2013,135, 16526

Contributed by Steven Bachrach.
Reposted from Computational Organic Chemistry with permission

Naphthalene, phenanthrene and pyrene are all planar aromatic compounds. How can substituted version be chiral, with the chirality present in the aromatic portion of the molecule? The answer is provided by Yamaguchi and Kwon.¹ They prepared peri-substituted analogues with the bulky adamantly group as the substituents. This bulky requires one adamantyl group to be position above the aromatic plane and the other below the plane, as in 1 and 2.

1

2

These molecules and two other examples were prepared in their optically pure form. B3LYP/6-31G(d) computations were performed on both of these structures (shown in Figure 1), but computations are a minor component of the work. These structures do show the out-of-plane distortions at C₁ and C₈, also apparent in the crystal structures. Computations of naphthalene and 1,8-dimethylnaphthalene show a planar naphthalene backbone, but i­-propyl substitution does force the substituents out of plane.

1

2

Figure 1. B3LYP/6-31G(d) optimized structures of 1 and 2.

These types of systems continue to subject the notion of “aromaticity” to serious scrutiny.

References

(1) Yamamoto, K.; Oyamada, N.; Xia, S.; Kobayashi, Y.; Yamaguchi, M.; Maeda, H.; Nishihara, H.; Uchimaru, T.; Kwon, E. "Equatorenes: Synthesis and Properties of Chiral Naphthalene, Phenanthrene, Chrysene, and Pyrene Possessing Bis(1-adamantyl) Groups at the Peri-position," J. Am. Chem. Soc. 2013,135, 16526-16532, DOI: 10.1021/ja407800e.

InChIs

1: InChI=1S/C30H36/c1-3-25-4-2-6-27(30-16-22-10-23(17-30)12-24(11-22)18-30)28(25)26(5-1)29-13-19-7-20(14-29)9-21(8-19)15-29/h1-6,19-24H,7-18H2
InChIKey=QNPJKZPPLCPHSS-UHFFFAOYSA-N

2: InChI=1S/C36H38/c1-2-27-4-5-28-6-7-30(35-15-21-8-22(16-35)10-23(9-21)17-35)34-31(14-29(3-1)32(27)33(28)34)36-18-24-11-25(19-36)13-26(12-24)20-36/h1-7,14,21-26H,8-13,15-20H2
InChIKey=DKRBDGNWYTWNHL-UHFFFAOYSA-N

This work is licensed under a Creative Commons Attribution-NoDerivs 3.0 Unported License.

Synthesis, Structural Analysis, and Properties of [8]Circulenes

Feng, C.-N.; Kuo, M.-Y.; Wu, Y.-T. Angew. Chem. Int. Ed. 2013, 52, 7791
Contributed by Steven Bachrach.
Reposted from Computational Organic Chemistry with permission

Circulenes are molecules where a central ring is composed of fused benzenoids. Corranulene can also be named [5]circulene and coronene is [6]circulene. In a previous post I discussed the topology of the circulenes. This earlier work suggested that [8]annulene 1 would have a saddle-shape. This hypothesis has now been confirmed with the synthesis of the substituted [8]circulene 2 by Wu and co-workers.¹

1

2

The x-ray structure does show a saddle geometry for 2. The central 8-member ring is tub-shaped, even more puckered that cyclooctatetraene (COT) itself, though the bonds in 2 are nearly of equal length. The bond lengths involving the central carbon atoms appear consistent with an [8]radialene-type structure.

The ωB97X-D/6-31G** optimized geometries of the parent compound 1 and the synthesized compound2 are shown in Figure 1. These computed structures are very similar to each other, along with being very similar to the x-ray structure of 2.

1

2

Figure 1. ωB97X-D/6-31G** optimized geometries of 1 and 2.
(Don’t forget that you can click on these structures – and any other structure on my blog – to interactively manipulate and visualize them, something worth doing here!)

The computed NICS(0) (at HF/6-31+G* – I would really rather have seen these computed with some density functional, preferably at ωB97X-D/6-31G**) values for the six-member rings of both 1 and 2 are negative, ranging from -8.9 ppm to -4.0 ppm, indicating aromatic character. The NICS(0) value at the center of the 8-member ring is +9.8 ppm in 1 and +12.2 ppm in 2. The authors argue that this value cannot discriminate the 8-member ring from that in COT (NICS(0) = 1.98 ppm, the expected value for a non-aromatic ring) and [8]radialene (NICS(0) = -1.2 ppm, also an expected value for a non-aromatic ring). However, they are silent on whether this might actually imply some antiaromatic character to the 8-member ring, which would be consistent with the equivalent bond lengths around the ring.

The authors note that there should be a second isomer of 2 resulting from a flip of the tub. Variable temperature NMR does not show any change in the spectrum, though with a different substituted [8]circulene they do see some coalescence, suggesting a large flipping barrier of at least 20 kcal mol^-1. A computational search for this flipping/inversion might be interesting as the transition state is likely to not be planar.

References

(1) Feng, C.-N.; Kuo, M.-Y.; Wu, Y.-T. "Synthesis, Structural Analysis, and Properties of [8]Circulenes,"Angew. Chem. Int. Ed. 2013, 52, 7791-7794, DOI: 10.1002/anie.201303875.

InChIs

1: InChI=1S/C32H16/c1-2-18-5-6-20-9-11-22-13-15-24-16-14-23-12-10-21-8-7-19-4-3-17(1)25-26(18)28(20)30(22)32(24)31(23)29(21)27(19)25/h1-16H
InChIKey=BASWMOIVIHXTRC-UHFFFAOYSA-N

2: InChI=1S/C96H80/c1-49-17-33-65(34-18-49)81-73-57(9)58(10)75-83(67-37-21-51(3)22-38-67)85(69-41-25-53(5)26-42-69)77-61(13)62(14)79-87(71-45-29-55(7)30-46-71)88(72-47-31-56(8)32-48-72)80-64(16)63(15)78-86(70-43-27-54(6)28-44-70)84(68-39-23-52(4)24-40-68)76-60(12)59(11)74(82(81)66-35-19-50(2)20-36-66)90-89(73)91(75)93(77)95(79)96(80)94(78)92(76)90/h17-48H,1-16H3
InChIKey=DEKWLSGHBADDAQ-UHFFFAOYSA-N

This work is licensed under a Creative Commons Attribution-NoDerivs 3.0 Unported License.

The Fold-In Approach to Bowl-Shaped Aromatic Compounds: Synthesis of Chrysaoroles

Myśliwiec, D.; Stępień, M. Angew. Chem. Int. Ed. 2013, 52, 1713
Contributed by Steven Bachrach.
Reposted from Computational Organic Chemistry with permission

Mysliwie and Stepian report on a new method for creating buckybowls.¹ The usual way had been to build from the inside outward. They opt instead to build from the outside in and have constructed the heterosubstitued bowl chrysaorole 1.

1

B3LYP/6-31G** optimizations reveal two conformers that are very close in energy: one has the butyl chains outstretched (1a) and one has the butyl arms internal or pendant (1b). These structures are shown in Figure 1. The depth of this bowl (1.96 Å) is quite a bit larger than in corranulene (0.87 Å). The agreement between the computed and experimental ¹³C and ¹H chemical shifts are excellent, supporting the notion that this gas phase geometry is similar to the solution phase structure. Though 1 is strained, 53.4 kcal mol^-1 based on B3LYP/6-31G** energies for Reaction 1 (which uses the parent of 1 – replacing the butyl groups with hydrogens), on a per sp² atom basis, it is no more strained than corranulene.

1a

1b

Figure 1. B3LYP/6-31G** optimized geometries of two conformers of 1.

Reaction 1

This new synthetic strategy is likely to afford access to more unusual aromatic structures.

References

(1) Myśliwiec, D.; Stępień, M. "The Fold-In Approach to Bowl-Shaped Aromatic Compounds: Synthesis of Chrysaoroles," Angew. Chem. Int. Ed. 2013, 52, 1713-1717, DOI:10.1002/anie.201208547.

InChI

1: InChI=1S/C54H45N3/c1-4-7-16-55-49-19-31-10-12-33-21-51-45-27-39(33)37(31)25-43(49)44-26-38-32(20-50(44)55)11-13-34-22-52-46(28-40(34)38)48-30-42-36(24-54(48)57(52)18-9-6-3)15-14-35-23-53(47(45)29-41(35)42)56(51)17-8-5-2/h10-15,19-30H,4-9,16-18H2,1-3H3
InChIKey=VUUJVETWVYQACL-UHFFFAOYSA-N

This work is licensed under a Creative Commons Attribution-NoDerivs 3.0 Unported License.

Bowl-Shaped Fragments of C70 or Higher Fullerenes: Synthesis, Structural Analysis, and Inversion Dynamics

Wu, T.-C.; Chen, M.-K.; Lee, Y.-W.; Kuo, M.-Y.; Wu, Y.-T. Angew. Chem. Int. Ed. 2013, 52, 1289-1293
Contributed by Steven Bachrach.
Reposted from Computational Organic Chemistry with permission

I have discussed a few bowl-shaped aromatics in this blog (see for example this and this). Kuo and Wu now report on a few bowls derived from C₇₀-fullerenes.¹ The bowl 1 was synthesized (along with a couple of other derivatives) and its x-ray structure obtained. As anticipated this polyclic aromatic is not planar, but rather a definite bowl, with a bowl depth of 2.28 Å. This is less curved than when the fragment is present in C₇₀-fullerene.

1

Interestingly, this bowl does not invert through a planar transition state. The fully planar structure 1pl, shown in Figure 1, is 116 kcal mol^-1 above the ground state bowl structure, computed at B3LYP/cc-pVDZ. Rather, the molecule inverts through a twisted S-shaped structure 1TS, also shown in Figure 1. The activation barrier through 1TS is 80 kcal mol^-1. This suggests that 1 is static at room temperature, unlike corranulene which has an inversion barrier, through a planar transition state, of only 11 kcal mol^-1. The much more concave structure of 1 than corranulene leads to the greatly increased strain in its all-planar TS. This implies that properly substituted analogues of 1 will be chiral and configurationally stable. Not remarked upon is that the inversion pathway, which will interchange enantiomers when 1 is properly substituted, follows a fully chiral path, as discussed in this post.

1
1TS	1pl

Figure 1. B3LYP/cc-pVDZ optimized geometries of 1, 1TS, and 1pl.

Reference

(1) Wu, T.-C.; Chen, M.-K.; Lee, Y.-W.; Kuo, M.-Y.; Wu, Y.-T. "Bowl-Shaped Fragments of C₇₀ or Higher Fullerenes: Synthesis, Structural Analysis, and Inversion Dynamics," Angew. Chem. Int. Ed. 2013, 52, 1289-1293, DOI: 10.1002/anie.201208200.

InChIs

1: InChI=1S/C38H14/c1-3-17-21-11-7-15-9-13-23-19-5-2-6-20-24-14-10-16-8-12-22-18(4-1)27(17)33-35-29(21)25(15)31(23)37(35)34(28(19)20)38-32(24)26(16)30(22)36(33)38/h1-14H
InChIKey=KLCLRPVBVWXTPN-UHFFFAOYSA-N

This work is licensed under a Creative Commons Attribution-NoDerivs 3.0 Unported License.

Synthesis, Characterization, and Computational Studies of Cycloparaphenylene Dimers

Xia, J.; Golder, M. R.; Foster, M. E.; Wong, B. M.; Jasti, R. J. Am. Chem. Soc. 2012, 134, 19709
Contributed by Steven Bachrach.
Reposted from Computational Organic Chemistry with permission

Nanotubes are currently constructed in ways that offer little control of their size and chirality. The recent synthesis of cycloparaphenylenes (CPP) provides some hope that fully controlled synthesis of nanotubes might be possible in the near future. Jasti has now made an important step forward in preparing dimers of CPP such as 1.¹

1

2

They also performed B3LYP-D/6-31G(d,p) computations on 1 and the directly linked dimer 2. The optimized geometries of these two compounds in their cis and trans conformations are shown in Figure 1. Interestingly, both compounds prefer to be in the cis conformation; cis-1 is 10 kcal mol^-1 more stable than trans-1 and cis-2 is 30 kcal mol^-1 more stable than the trans isomer. While a true transition state interconnecting the two isomers was not located, a series of constrained optimizations to map out a reaction surface suggests that the barrier for 1 is about 13 kcal mol^-1. The authors supply an interesting movie of this pseudo-reaction path (download the movie).

cis-1	trans-1
cis-2	trans-2

Figure 1. B3LYP-D/6-31G(d,p) optimized geometries of the cis and trans conformers of 1 and 2. (Be sure to click on these images to launch a 3-D viewer; these structures come to life in 3-D!)

References

(1) Xia, J.; Golder, M. R.; Foster, M. E.; Wong, B. M.; Jasti, R. "Synthesis, Characterization, and Computational Studies of Cycloparaphenylene Dimers," J. Am. Chem. Soc. 2012, 134, 19709-19715, DOI: 10.1021/ja307373r.

InChIs

1: InChI=1S/C106H82/c1-5-13-79-21-9-17-76-29-37-85(38-30-76)95-59-63-98(64-60-95)103-71-69-101(82(16-8-4)24-12-20-77-27-35-84(36-28-77)90-51-55-94(56-52-90)91-45-41-86(79)42-46-91)73-105(103)99-65-67-100(68-66-99)106-74-102-70-72-104(106)97-61-57-88(58-62-97)81(15-7-3)23-10-18-75-25-33-83(34-26-75)89-49-53-93(54-50-89)92-47-43-87(44-48-92)80(14-6-2)22-11-19-78-31-39-96(102)40-32-78/h5-16,21-74H,1-4,17-20H2/b21-9-,22-11-,23-10-,24-12-,79-13+,80-14+,81-15+,82-16+
InChIKey=WFVBBCVHFBTQRK-VPGVYKRGSA-N

2: InChI=1S/C100H78/c1-5-13-75-21-9-17-72-29-37-81(38-30-72)91-59-63-94(64-60-91)97-67-65-95(78(16-8-4)24-12-20-73-27-35-80(36-28-73)86-51-55-90(56-52-86)87-45-41-82(75)42-46-87)69-99(97)100-70-96-66-68-98(100)93-61-57-84(58-62-93)77(15-7-3)23-10-18-71-25-33-79(34-26-71)85-49-53-89(54-50-85)88-47-43-83(44-48-88)76(14-6-2)22-11-19-74-31-39-92(96)40-32-74/h5-16,21-70H,1-4,17-20H2/b21-9-,22-11-,23-10-,24-12-,75-13+,76-14+,77-15+,78-16+
InChIKey=HOODCSIDKUJYKE-XJQPCHFNSA-N

This work is licensed under a Creative Commons Attribution-NoDerivs 3.0 Unported License.

Chemical Networks (Triple Header!)

1. Rewiring Chemistry: Algorithmic Discovery and Experimental Validation of One-Pot Reactions in the Network of Organic Chemistry 
2. Parallel Optimization of Synthetic Pathways within the Network of Organic Chemistry Chemical
3. Network Algorithms for the Risk Assessment and Management of Chemical Threats 

A back-to-back-to-back (!) set of three papers in Angewandte Chemie from Bartosz Grzybowski and co-workers. All three articles concern the development of Chemical Networks and their application in synthetic chemistry, of which more later. It is often said that the realm of synthesis is both art and science, however, the wealth of empirical observations made over centuries of making molecules underpin the field. As Grzybowski remarks here, “it is simply beyond cognition of any individual human to understand and analyze all this collective chemical knowledge”, and most chemists already search online synthesis databases to perform individual steps, but perhaps the role of automated computational synthetic route selection, and reaction design is set to grow? Also see Dean Tantillo's recent post on CCH.

Grzybowksi’s group has constructed a network of organic chemistry (NOC) from reactions in the chemical literature since 1779 until present day: reactants and products are represented by nodes in this graph and known chemical interconversions by edges. From this NOC containing seven million reactions, the first paper of the series seeks to discover new ways of performing consecutive reactions in the same vessel (so-called “one pot” reactions). From known reactions that interconvert A to B and B to C, the authors have coded filters that check for compatability between solvents, reagents, catalysts etc so that the two steps may be performed in the same reaction vessel, thus creating a novel way to prepare C from A in one step. Typically synthetic organic protocols are the result of much tinkering and optimization studies in the lab: in contrast the NOC predictions have yield a number of two, three and four step one-pot reactions that give moderate to good yields without any human optimization.

In the second paper the group turn their attention to designing “optimal” reaction pathways to synthesise a given target molecule. Again the NOC is used, this time to propagate backwards from the target via an initial synthetic plan to starting materials. A Metropolis Monte Carlo algorithm is used to randomly sample alternative routes in order to minimize a penalty function associated with the cost of performing each step. Impressively this approach has been used already by a synthesis company to reduce their costs. Additional costs such as waste disposal or energy costs associated with heating/cooling are undoubtedly important for chemistry on the process scale, and perhaps these could be incorporated in future implementations of the optimization.

The third application of a chemical network considers the synthesis of chemical warfare agents. Reaction networks are explored starting from commonly available household chemicals. Thankfully the paper is careful not to disclose any of the synthetic steps involved, and the authors propose strengthening existing regulation of substances by not only regulating single molecules but also combinations of reagents that have been ranked according to game theory as more likely to be used.