Recent trends of Flow Chemistry #memo #chemistry #technology

I enjoyed reading the article in my lunch break. Researchers at Abbie published nice review about the flow chemistry in the pharmaceutical industry. The URL is below.

The review describes wide range of the flow chemistry from large scale synthesis to lab scale synthesis.

Flow chemistry can run the reaction under high temperature and high pressure reaction conditions. Photo redox reaction is also available. I like rearrangement reactions because the reaction is atom economic and stereo selective I feel it is elegant. But theres reactions often requires high temperature condition. In scheme 10 shows example of Overman Rearrangement with flow. The example shows >95Kg scale synthesis in 84hr! Continuous synthesis is powerful tool for production.

One of the strong point of flow reaction is flush chemistry that can conduct reaction in very short time. It means the method can handle unstable(reactive) intermediates. In scheme 15 shows flow synthesis example of Eribulin intermediate. You know Eribulin is a laboratory-made form of halichondrin B, a substance. It has very complex structure. The researchers conducted DIBAL-H reduction of ester to aldehyde and then conducted julia coupling type anion addition. To use flow reaction reaction temperature was raise up from -70 deg to 10 deg.

There many examples are described in the article and scheme35 Flow diazomethane chemistry seems very useful. Diazomethane is useful reagent but sometime it is difficult to use in lab for safety reason so TMS-diazomethane was used. But TMS-diazomethane is not cost effective. The Scheme35 shows example tube-in-tube reactor. What is tube-in-tube reactor? I would like to draw the image below. The inner tube is made with teflon AF-2400 which is gas-permeable tube. The tube can through diazomethane only so outer layer trap CH2N2 and generate pure CH2N2/THF solution. I have not known the technology. It is cool. Reader who has interest, pls read the article.

___________________________________________________ outer tube
THF>>>>>>>> CH2N2/THF
___________________________________________________ AF2400
Diazald+KOH >>> CH2N2 aq + side products >> side products
___________________________________________________ AF-2400
TFH>>>>>>> CH2N2/THF
___________________________________________________ outer tube

Above examples are production. BTW, how about parallel chemistry?

According to the article, Abbie was developed SWIFT (synthesis with integrated flow technology, nice naming sense!). The system integrates flow reactor and HPLC/MS and be able to synthesize 6 pure compound per hour in 10 ~ 20mg scale! It is very productive I think. Even if I use parallel reactor such as miniblock, it is difficult to synthesis pure compounds such speed. Fig 24 shows SWIFT platform. It seems not so large. And there are some examples continuous synthesis and assay cycles. In Japan there are few examples of parallel chemistry with flow I think.

If we can run DMTA cycle within few hours, what will be task of medicinal chemists.


Change properties of approved oral drugs

When I learned drug discovery long time ago, I read the article about Role of five which is a rule of thumb to evaluate druglikeness.
You can read nice review about the druglikess scores in following URL.
( Written in Japanese ;-) )
View story at

By the way, recently there are many articles which are describing about beyond the Role of five . I read an article published from JMC, reported by researcher of NIB.

The author analyzed the change over time of properties such as MW, logP, HBA, HBD, TPSA and NumAromaticRing.
In page B, the author reported comparison between experimental and calculated (StarDrop, Pomona, Moka, Crippen) value of LogP which is determined in Novartis. And the experiments shows that StarDrop gave the lowest variability between measured and calculated logP. Most of tools overestimate in high log P area but stardrop does not. It worth to know that clogp is largely affected by computational tools.

In Table5 shows analysis of FDA approved oral NCEs from 1998 to 2017. Mw, RotBm and #ArRNG showed significance.
For example 90th percentile of Mw is 571.0 compared to 1997 90th percentile 470.3. It dramatically increased!
And also 90th percentile of #ArRNG is 4 compared to 1997 90th percentile 3.
Mw is changed dramatically I think. Fig 6 shows Number of ratio of MW in approved drug in each time period. This figure shows tendency to decrease of ratio of the drug which has MW <= 500.
By the way, HBD is not changed.

The author discussed about why size (MW) is matter and described that higher molecular weight is not be indicator of druglikeness but be estimator of synthetic accessibility.
To synthesize molecule with high molecular weight, it needs more building blocks and / or longer synthetic steps. So it is needed more effort to SAR expansion in medicinal chemistry projects.

Ro5 is easy to understand and reasonable role for medicinal chemistry but it is changing. Recently there are many modalities for drug discovery.
Changing role is needed long time and many efforts…..
More details are described in the article. It is informative for me.

Applicable Domain on Deep Neural Networks #JCIM #chemoinformatics

I read interesting article from JCIM.
Dissecting Machine-Learning Prediction of Molecular Activity: Is an
Applicability Domain Needed for Quantitative Structure−Activity
Relationship Models Based on Deep Neural Networks?

URL is below.

The pros of DNN is feature extraction. And there are many articles which use DNN for molecular activity prediction. BTW, is it true that DNN is outperform any other machine learning methods?

The authors of the article analyzed the performance of DNN. They used ECFP4 as an input feature and predicted biological activities extracted from CHEMBL DB.
Their approach was reasonable, they built model with training set and check the performance with test data and evaluate RMSE in several layers which are defined by molecular similarity. Layer1 means that dataset is similar to training data and Layer6 means that dataset is not similar to training set.

They analyzed performance of predictive method such as KNN, RF and DNN and their analysis revealed that DNN showed similar performance with RF and KNN. And also Fig 5 shows that DNN can not predict objective value when query molecule is not similar to training set.
It indicates that DNN does not learn feature of molecule from finger print but learned pattern of fingerprint.
More details are described in this article.
In the real drug discovery project, MedChem sometime designs not seed similar compounds. For the chemist, this is not so special. But it is difficult point for AI to learn sense of MedChem.

Biological activity prediction is challenging area I think, And ECFP is still de fact standard for chemoinformatics. I would like to develop new concept of molecular descriptors.

(new?) medchem tool box for compound synthesis

This mini perspective shows recent progress of the direct C-H alkylation with Alkyl Sulfinates.

There are many heterocyclic moieties such as pyridine, pyrrole etc. in drug like molecules.
The C-H alkylation reaction of heterocycles is useful but difficult to conduct it under mild reaction conditions.
So mild and universal reaction is very attractive for chemists I think.

In the article, Phil S. Baran’s group reported many examples of their developed reaction.
They used alkyl sulfinates as a radical precursors and developed many kind of commercial sulfinates, fluoroalkyl, heterocyclic, alkyl, aromatic and linker-type.

Surprisingly most of the reactions proceed room temperature. Also they shows reactivity guid lines, it is very practical!

And also this reaction can apply not only early stage of synthesis but also late stage of synthesis.
It means that this is very specific reaction.
I had few successful cases C-H alkylation with sulfinate…. I would like to use the reaction condition when I have chance.

Similar but not similar compounds….

Some months ago I wrote a blog post about biased ligand. It is still exciting area for me.

BYW, the post is about Rexulti developed by Otuka Pharmaceutical.
And the drug approved 19th Jan. 2018 in Japan for schizophrenia. Just by looking, Rexulit has very similar structure to Abilify.
I interested the patent strategy and got some U.S. patent from google patent.


U.S. patent number of Abilify is 5,006,528 filed in oct. 20, 1989 “CARBOSTYRIL DERIVERTIVES”.
Common structure is below.

R in the formula is limited substituted benzene.
In this patent, there are many in vivo data but no in vitro data.

Let’s move to next patent, rexulti.
I found following U.S. patent.
The claimed structure is below.

In the figure, Q is calbostyril moiety. Main difference is monocyclic benzene and benzothiophene.
In this patent, inventor differentiate competitors patents from pharmaceutical or structural side.
Let’s see background art!
WO2004/026864A1 discloses that acarbostyril derivative represented by the general formula.

It seems too similar I think. But they differentiate from biological activity.
However, there is no description in WO2004/026864A150 that carbostyril derivatives described in the document have D receptor partial agonist activity,5-HT2 receptor antagonist activity, a receptor antagonist activity and serotonin uptake inhibitory activity together and have a wide treatment spectrum.
Hmm. OK CNS area is very complex. It will be difficulty point for drug discovery but also will be chance.

Next patent WO 2005/019215A1 disclose the following formula.

The claimed structure is also similar. But the patent does not disclose the structure of rexuti.
However, WO 2005/019215 A1 does not specifically disclose the compounds of the present invention

There are lots of patents about Abilify and Rexulti. In this industry patent strategy is very important because finding new drug is needed long time and efforts.

This post shows only very easy part these drugs but I feel very difficult and interesting point of CNS area drug discovery.

Read a letter.

To design molecule, I care about not only biological activity but also molecular properties.
Because if chemist do molecular design only focused on biological activity, it will produce many problems.

I read short letter of J. Med. Chem. lett. ASAP article.
The letter shows some interesting data.
I often use LE, LLE but not used PFI before.
GSK uses PFI (property forecast indices) to asses the probability of development risk.
PFI is calculated by sum of Chrom LogD + HAr rings. And They shows probability of risk as traffic light colors. I like the analyse.

And there are some papers that analysed the impact of aromatic ring on compound developability.
LogP, Num of aromatic rings is good indices because it’s easy to understand and big impact for drug likeness.
So, I think PFI is interesting parameter.
And I think the traffic light colors Green(go!), Yellow(careful), Red(Stop) is very easy to understand I like it.

Another interesting data, They summarised mean of vitro activity and property changes in optimizations reported in major medchem papers in table1.
The author shows there was significant difference in LLE between two groups. One group assessed lipophilic influence in design. And another group did not.
If the data is true, I want to know the origin of Yes group is pharmaceutical companies or academia or both.
I often fall in a dilemma, need more activity but increase lipophilicity to increase activity, it maybe not good for another properties.
Am I not good medicinal chemist ?
There are a lot of method to design molecule i.e. SBDD, FBDD, CADD, or HTS etc. I have to use these tools and organise data and design molecule more efficiently.
Need study more and more…