Which Compounds Should We Synthesise Next? June 2015

[alintheopen]

In the next week or so we need to decide on OSM's synthetic targets for the next few months.
Blog post here (http://malaria.ourexperiment.org/uri/7fe) provides a summary of some of the data and some suggested molecules. Suggestions/comments/criticism welcome as always so please comment here, below the blog or to @O_S_M on Twitter.

[mattodd]

Lots of options. My sense is, first of all, that the low-hanging fruit are the compounds in Part 2 - the substituted pyridines. Compounds A-C are still interesting and would complete a good exploration of the Northeast.

We could use a synthetic plan for compounds with the altered core (O and P). I might ask whether previous synthetic routes from the CRO are available. I wonder about blocking that position (the place you have a dot) on MMV669846 with a methyl etc, otherwise we may have the same clearance issues.

I'll set up a separate issue for scheduling the next open planning meeting.

[drc007]

A few thoughts, feel free to ignore.
J,L,M are hemiacetals and will not be stable, similarly I and N are aminals.
Why not make the benzyl ethers at position X,
You can also add alpha substitution on the benzyl ethers to investigate conformational constraints.
Do you have measured LogP on any analogues? You need to be careful basing your design criterion for lipophilicity on only calculated values.

[mattodd]

Quite so on the acetals/aminals.

Benzyl ethers at position X is something we tried (I think you previously suggested this) and the potencies weren't great. e.g. http://malaria.ourexperiment.org/biological_data/11216/Evaluation_of_Latest_Series_4_Analogs_in_Ether_and_Amide_Series.html
..but alpha substitution is not something we looked at. Any suitable commercially-available compounds we could just add in?

We have a little logD: http://malaria.ourexperiment.org/biological_data/8408/post.html but we are intending to get more in this series to be sure, yes. Ideally we'd have had more activity in this latest set.

[drc007]

I can only see one example? MMV675959 450 nM. The SAR might be different to phenethyl analogues.
Might be better to make the benzyl ether analogues with 2,3 and 4 monochloro substitution.
Alpha substitution can be achieved via the ester http://www.sciencedirect.com/science/article/pii/S0960894X97101184, apologies for quoting one of my own publications ;-)

[mattodd]

;) Sure - those are good suggestions. From the wiki there's also the
transposed ether MMV670946 at 521 nM and the naphthyl compound MMV672688.
Also note the interesting phenyl (rather than benzyl) MMV669784: not great
potency but clearance not bad. I guess that is our compromise here. Longer
chains provide improved potency but introduce the metabolic liability.

On 17 June 2015 at 00:02, Chris Swain [email protected] wrote:

I can only see one example? MMV675959 450 nM. The SAR might be different
to phenethyl analogues.
Might be better to make the benzyl ether analogues with 2,3 and 4
monochloro substitution.
Alpha substitution can be achieved via the ester
http://www.sciencedirect.com/science/article/pii/S0960894X97101184,
apologies for quoting one of my own publications ;-)

—
Reply to this email directly or view it on GitHub
#301 (comment)
.

MATTHEW TODD | Associate Professor
School of Chemistry | Faculty of Science

THE UNIVERSITY OF SYDNEY
Rm 519, F11 | The University of Sydney | NSW | 2006
T +61 2 9351 2180 | F +61 2 9351 3329 | M +61 415 274104
E [email protected] | W
http://sydney.edu.au/science/chemistry/research/todd.html | W
http://opensourcemalaria.org/

CRICOS 00026A
This email plus any attachments to it are confidential. Any unauthorised
use is strictly prohibited. If you receive this email in error, please
delete it and any attachments.

[sdebbert]

I agree on the low-hanging fruit. In class this May, I had students make the 2- and 3-pyridylhydrazones, both of which crashed out beautifully from ethanol (esp. the latter).

In general, hydrazone formation went well (aliphatic and aromatic aldehydes, about ten in all). For the oxidative cyclization, I had half the class use PIDA and the other half use Chloramine-T. The latter didn't work at all -- an old, bad bottle of the oxidant, I think -- while the former gave the desired product relatively cleanly. The final nucleophilic aromatic substitution step, though, cratered; it was a much faster reaction than I anticipated, and I think it degraded the product fully between lab sessions. I just presented this work at the Northeast Regional ACS meeting in Ithaca, NY.

This summer, I and my research student Charlie Martin intend to:
a) see if we can take some of the synthetic intermediates from class to the final products,
b) post the data we get, and
c) explore the use of C-H activation/directed metallation chemistry in derivatizing the parent triazolopyrazine core more directly (i.e., post-cyclization derivatization).

[alintheopen]

Hi all,

apologies for my mistake with semi-aminal/acetal - I had meant to include a
carbon chain substituted with a polar group. I'm interested in Chris' ideas
for the benzyl ethers, will check CLogP. Also agreed on Log P measurements
something we need to do pretty soon.

On Wed, Jun 17, 2015 at 4:43 AM, Stefan Debbert [email protected]
wrote:

I agree on the low-hanging fruit. In class this May, I had students make
the 2- and 3-pyridylhydrazones, both of which crashed out beautifully from
ethanol (esp. the latter).

In general, hydrazone formation went well (aliphatic and aromatic
aldehydes, about ten in all). For the oxidative cyclization, I had half the
class use PIDA and the other half use Chloramine-T. The latter didn't work
at all -- an old, bad bottle of the oxidant, I think -- while the former
gave the desired product relatively cleanly. The final nucleophilic
aromatic substitution step, though, cratered; it was a much faster reaction
than I anticipated, and I think it degraded the product fully between lab
sessions. I just presented this work at the Northeast Regional ACS meeting
in Ithaca, NY.

This summer, I and my research student Charlie Martin intend to:
a) see if we can take some of the synthetic intermediates from class to
the final products,
b) post the data we get, and
c) explore the use of C-H activation/directed metallation chemistry in
derivatizing the parent triazolopyrazine core more directly (i.e.,
post-cyclization derivatization).

—
Reply to this email directly or view it on GitHub
#301 (comment)
.

[mattodd]

That'd be awesome @sdebbert - to get to the final products, but also to learn what other hydrazones you've been using so that we don't duplicate. As for the late-stage derivatisation, that was something @tscmacdonald looked at a little, via halogenation, but he didn't have time to finish. Might be an option for you to look at - short summary here: http://openwetware.org/wiki/OpenSourceMalaria:Triazolopyrazine_%28TP%29_Series#Alternative_Routes_to_the_Triazolopyrazine_Core

[wvanhoorn]

I am new to this joint effort so let me know if the below is appropriate. The company I work for has developed algorithms to automate compound design in hit finding and hit to lead optimisation. A colleague sent me a link to your latest blog post and I have tried to apply some of our technology to generate some quick wins by exploiting Cores and RGroups that have been used before.

• 128 compounds already made and assayed were copied from Malaria Moleules spreadsheet on 16 June 2015.
• Two cores were mentioned on the OSM blog. These cores were used to deconvolute the 128 structures in 107 unique R1/R2 groups. Enumerating the two cores with the 107 R1/R2 yields 2760 virtual compounds. Of these 106 are already known leaving 2554 new structures. By InChi matching it could be established that compounds E and O from the blog are contained in the set.
• A Bayesian activity model 'MMV_NB' was derived from the 119 compounds with a value for 'PfaI EC50 uMol' based on ECFP_6 fingerprints. A quick leave-one-out validation showed that it has moderate predictive power.
• The 2554 structures were filtered together with the 12 suggestions from the blog where structures were provided in a machine readable format (MK-001 to MK-012). The filter was arbitrarily set to MMV_NB > 5 and QED > 5 leaving 276 structures. No structural filters were applied since all substructures have been seen before and are therefore assumed to be acceptable.
• A subset of 10 structures was selected by an as yet proprietary active learning algorithm. Ten was arbitrarily chosen, it can be set higher or lower if required.
• All selected designs share the same core, it is trivial to balance the designs more.
• The graph below show the MMV_NB score vs QED for the 2666 virtual compounds. The structures failing the QED or MMV_NB cut-offs are coloured grey. Structures passing the filters have been Pareto ranked and are coloured by Pareto front (red to blue = most to least attractive). The active learning selection is coloured green and finally MK-xxx are represented as triangles amd compounds E and O as stars.
  

Google sheet with 10 designs

[mattodd]

Very interesting analysis and suggestions, @wvanhoorn . Three quick points before more specific discussions:

1. The top 10 look to be all imidazopyrazines, rather than triazolopyrazines. Is that right? Certainly these are of interest, but as part of a mixed set of target structures.

2. We have slightly deprioritised the amide sub-series (i.e. carbonyl-NH attached to pyrazine) because of some hERG and aldehyde oxidase data. For this next set we'd like to avoid those, I think. We'd been focussing on ethers (many of which are in your top 10) since we know amines and thioethers are no good. Any way to bump those up the Top 10 list?

3. Can we enforce a cLogP maximum of 3.5?

[alintheopen]

Very nice thanks @wvanhoorn, I am also intrigued by selection of imidazopyrazines over triazolopyrazines.

@drc007 are these (below) the benzyl ethers that you had in mind? Of concern is CLogP values >5 but we could make a couple for sure.

[drc007]

Hi,

Yes those are the compounds.

On 18 Jun 2015, at 05:58, alintheopen [email protected] wrote:

Very nice thanks @wvanhoorn https://github.com/wvanhoorn, I am also intrigued by selection of imidazopyrazines over triazolopyrazines.

@drc007 https://github.com/drc007 are these (below) the benzyl ethers that you had in mind? Of concern is CLogP values >5 but we could make a couple for sure.
https://cloud.githubusercontent.com/assets/2626599/8224449/7f03f73c-15ca-11e5-9af4-31acbbd33fa0.png
—
Reply to this email directly or view it on GitHub #301 (comment).

[wvanhoorn]

@mattodd

1. The active learning algorithm balances novelty with predicted activity (exploration vs exploitation). Since far fewer imidazopyrazines have been synthesised than triazolopyrazines there is a strong bias towards the former esp since the Bayesian predictions are favourable. I have rerun the algorithm forcing it pick structures with a specified core. Below are two sets of 10 compounds based on each core.
2. The easiest way to remove the amide compounds is filtering them out using a substructure filter which has been done
3. I don't have access to ClogP but I have now filtered by AlogP < 3.5 (which is more or less the same thing)

2 x 10 designs

[mattodd]

Hi @wvanhoorn - thanks again! Three tweaks/questions:

1. The dataset has been updated in the last 2 days with some compounds containing the Pyr ring on the triazole. They're inactive. This might alter the suggestions if you ran this again?

2. Could you remove the compounds MK-001 to MK-012 for now as these are suggestions, which may be good suggestions but we've no data on them yet.

3. The structures proposed are obviously hybrids of knowns. Is there a way to introduce functional groups or substructures that are similar to those we've tried (and were active) but still new to the series? Perhaps those that have shown usefulness in other medchem programs. If not, don't worry, but just wondering.

[wvanhoorn]

Hi @mattodd - you are welcome.

1. I did use a recent set including the inactive pyridines when I reran the calculation. So why are they still there in the designs? Firstly, many of the virtual compounds have higher Bayesian scores than the pyridines but also an AlogP > 3.5 so they were filtered out. Secondly, the pyridine-containing compounds still have high model scores since for all pyridines there is only one (inactive) example in the training set, i.e. an n=1 observation with low statistical weight. The pyridine contributes negatively to the score but not enough to bring the overall score down. See the structure below where the atoms are coloured by contribution to Bayesian score (red = positive = favourable, blue = negative = unfavourable). The pyridine is blue, but less intense than the red parts elsewhere in the molecule.
   
   After all filters have been applied only 58 triazolopyrazines remain. Out of this set, the active learning algorithm favours the pyridines because of the low statistical weight, more sampling is required to firm up on the observation they are inactive. Below is the graph of model scores vs AlogP with the cut-offs, only compounds from the top left quadrant were eligible.
   
   I would propose to increase the AlogP. Unshielded (no ortho substituent) pyridines run the risk of P450 inhibition, it's maybe a good idea to remove these altogether.
2. These compounds were only run through the model to see where they would rank compared to the interpolated structures (and to check if there was any overlap). They were not used to construct the model.
3. My colleague Jérémy Besnard has applied Matched Molecular Pairs transformations on selected highly active triazolopyrazines to see if close-in analogues could be generated. The good news is that there are lots, the bad news is also there are lots, this requires a bit more time to process. I will try to see if there is some time this weekend, otherwise it will be next week.

[alintheopen]

Thanks @wvanhoorn! Just a quick question, there is definitely more than one inactive pyridine and they should be on the G Sheet (http://malaria.ourexperiment.org/uri/76d). Is the 'training set' a smaller set of data than the Gsheet data? Cheers

[wvanhoorn]

@ALL,
Sorry for the confusion: the training set that was used does contains three inactive unsubstituted pyridines (MMV675961, MMV675962, MMV675949). These three contain one example each of ortho/meta/para pyridine, hence the use of the singular when discussing one of them in more detail.

Note there are also two substituted pyridines which are active (MMV670936 and MMV688895). In the final filtered set (top left quadrant in the Alog/MMV_NB graph in the previous post) there are 27 compounds containing a pyridine connected to the triazole. Of these, 23 are unsubstituted and 4 contain the CF3 in the 2 position as seen in MMV670936 or MMV688895. Because this substituted pyridine has been seen more often in the training set (only twice, still one more than each of the unsubstituted pyridines) and is far less prevalent in the final filtered set (4 vs 23) the active learning algorithm picks the unsubstituted pyridines first.

I think there are two approaches that could be taken:

1. Increase the AlogP threshold and/or lower the model score threshold to get a larger pool of compounds to pick from. For instance an increase of the AlogP cut-off to 4.5 more than doubles the pool. Lowering the model threshold will add more speculative compounds to the mix. To prevent the unsubstituted pyridines they can be filtered out explicitly just like the amide linkers. MK-001 to MK-012 would then also be back in contention.
2. Ditch the active learning selection and just pick the compounds with the hightest model score. The balance of exploitation vs exploration will be then 100% exploitation. Given that all of these compounds are interpolated (reassembled known Core/RGroups) this may be more appropriate anyway. Novelty can be brought in separately by the RGroups my colleague has derived.

My preference would be a combination, increase the AlogP threshold to 4.5 and exploitation by model score. Below is the top 25 when taking this approach. To widen the selection the QED filter was turned off as well, some structures may be less attractive. I don't know how many you have resources for to make, I have added the full list as gsheet just in case. When I have time I will have a look at the novel RGroups.

Full set as gsheet

[wvanhoorn]

As promised some time ago here are some complementary designs, i.e. compounds constructed using RGroups not seen before. They were derived by taking matched molecular pair transformations from ChEMBL and applying them to the existing RGroups. Extensive filtering has been applied to remove unstable or reactive structures, too many or too large new rings, etc. This removes most but not all 'odd' structures. Synthetic accessibility was not assessed and may be less than ideal. The 20 structures below are the top 10 by model score followed by 10 complementary structures picked by our active learning algorithm. The latter picks up more novel structures which may include some less desirable ones.

gsheet

[alintheopen]

I found two patents in SciFinder with synthetic schemes for formation of the 'new core'

6-chloropyrazin-2-amine (InChI=1S/C4H4ClN3/c5-3-1-7-2-4(6)8-3/h1-2H,(H2,6,8)) has 239 commercial sources listed on SciFinder and is reasonably priced. Some investigation needed as none of our usual suppliers listed.
http://worldwide.espacenet.com/publicationDetails/originalDocument?CC=WO&NR=2012015723A1&KC=A1&FT=D&ND=3&date=20120202&DB=EPODOC&locale=en_EP

[alintheopen]

Before ordering any SM, we will attempt to synthesise by reaction of 2,6-dichloropyrazine with ammonia.

[tyzhengsydney]

Hi, I am Tianyi. I am a student on exchange from Nanjing University and I am working with Alice on the lab. We try to make 6-chloropyrazin-2-amine under the same conditions used to make 2-Chloro-6-hydrazinylpyrazine (AEW 85-7), but replacing hydrazine hydrate with ammonia (http://malaria.ourexperiment.org/uri/810). The TLC so far suggests only starting material but we will confirm today. I did literature search and found a patent describing synthesis of 6-chloropyrazin-2-amine using the same starting material but with aqueous solvent at 100 degrees in a sealed tube. We will try that today.http://worldwide.espacenet.com/publicationDetails/originalDocument?CC=WO&NR=2013068755A1&KC=A1&FT=D&ND=&date=20130516&DB=&&locale=en_EP

[ghost]

Hi All:
Further to the suggestion [http://malaria.ourexperiment.org/the_osm_blog/11567/Series_4_Suggestions_from_Mrinal_Kundu.html], here I post some additional compounds for the discussion on 27th.

Feel free to comment and look forward to the discussion.

[ghost]

Its nice to see that there are plenty of suggestions from the members - thank you all.
Some molecules have free -OH groups, based on the lower cLogP and also good potency shown by the earlier analogues; considering the metabolism/high clearance for these, it may be better to replace -OH by -OMe.
Moreover a few compounds contain anilinic moiety - need to be cautious w.r. to potential toxicity. Also the presence of highly basic NH2/ NHMe/ NMe2 groups refers to cationic amphiphillic structures [CAD] those eventually cause phospholipidosis and potential hERG liability [@wvanhoorn, 19th, 22nd & 29th June comments].

[ghost]

Following are some humble comments on the post [having the following link] for the discussion on 27th: http://malaria.ourexperiment.org/the_osm_blog/11565/What_Should_We_Synthesise_Next.html

1. Remaining 'Top Ten' Compounds:

The recent results from OSM-S-275 & 276 are not surprising; as I had earlier commented [https://plus.google.com/communities/105318249049322037798], changing 4-Cl/4-OCHF2-Ph to either THP or N-linked Piperidine], there is a drastic loss in potency which suggests the need of pi-pi interaction/aryl group at that position.

In that context, remaining targets A & B may not add any advantage at this point. Similarly some compounds from @wvanhoorn [19th & 22nd June comments].
The target C may also be non-beneficial.

1. Exploration of substituted pyridines at X and or Y:

Compounds D & E are attractive and worth making - in similar line, MK-013 might be interesting too, which is an analogue of E with a assumption that this will have lower clearance.

1. Polar Benzylic Groups:

Proposed targets F-H have unsubstituted phenyl ring. Comparing MMV 672687 with OSM-S-279, we see decrease in potency by 2 folds. So, if we keep in mind the data for MMV670947, we may get reasonable affinity for G.

On the other hand F & H have basic amine groups [few examples also from @wvanhoorn, 19th, 22nd & 29th June comments] good to be avoided
Regarding I & J and L, M & N [from: Additional substitution on side chain X], team already have discussed.

1. Modification to the core:

Compound P may not be stable - we have F atom attached to carbon adjacent to N which is labile.

Feel free to ignore these comments.

In going forward, considering the data obtained so far [a few good potency, reasonable ADME, that too in vivo efficacy for 2 compounds from this series], we probably need to focus more to optimize to improve on the missing balanced criteria in a single compound [some thoughts as in MK-001 to 018 for discussion]; of course, it is certainly useful to allocate some time to get additional compounds to further explore the SAR.

Best regards.

[drc007]

Looking through the compounds made to date, I notice that there are no examples where the amide linker (e.g. MMV670944 or MMV668958) has been reversed. There are a couple of amine and asulphonamide linker which would suggest the chemistry should be accessible?

[ghost]

Hi Chris,

MMV669103 [reverse sulfonamide] is not promising [10 uM], though it is not a 'head-to-head' comparison to either of MMV670944 or MMV668958.
One direct analogue, like the one suggested, can be made and checked.

We may keep in mind that, in the case of amide series, C=O of the amide is probably at the 'same place' as the O-atom of the ether series and those are presumably acting as HBA!

Reversing the amide position, we will make this 'HBA' one bond away - will be interesting to see how this is tolerated.

Best regards.

[drc007]

I'm wondering if there is a h-bonding interaction. If you compare the following analogues they all seem to have very similar activity.

[drc007]

It might be worth investigating various linking groups, reverse amide, ketone, sulphone, sulphoxide?

[wvanhoorn]

@drc007 Just a question about the proposed H-bonding, the three compounds shown differ in their linkers: one with a donor and acceptor, one with a (weak) acceptor and one without any donors/acceptors. Does this not imply that H-bonding is not important?

[drc007]

Certainly a reasonable conclusion, this might mean we can use the linker to change the physicochemical properties of the molecules, the electronics of the rings to alter metabolism etc.

[ghost]

Good points drc007 and wvanhoorn - appreciate that.
I did check the comparison; moreover, I also considered the potency data of MMV670243, MMV669304 and MMV669027.

So, my comments on HBA was indeed a curiosity as I said, it will be interesting to see the effect while making the reverse amide suggested above. Team can thus decide on synthesizing a few direct analogues first and depending on the result can proceed accordingly.