Paper Explained: Using thermodynamically controlled networks to assess molecular similarity

I always wanted to do this kind of blog post. The idea is to explain a chemistry paper without jargon and just a little bit of chemistry. Let’s see if I can mange to do it:

Today I will (try to) explain our recent paper: Systems chemistry: using thermodynamically controlled networks to assess molecular similarity. It’s an open access paper, you are more than welcome to download and share it.


Why this research?

The similarity concept is quite hard to explain, and it can easily go into a philosophical discussion. Remaining on the scientific approach, take a look at these two molecules:




They look similar, isn’t it? But when you smell them the upper one will smell like coconut, while the bottom one like peach. That is quite a difference. This is because we have thousand of receptors in our body that can differentiate between these two molecules. This is a key point, it’s not only one receptor doing all the job, but it’s rather a network of receptors.

Medicinal chemistry and drug development are the main targets of this kind of research. We already have a number of working drugs, now we want to find some effective new molecules. Think for example at the antibacterial drugs, most of the standard one are ineffective today and we need new ones. So far, most of the pharmaceutical industries screening is based on computational approach.

We wanted an experimental setup with test molecules and receptors.The goal of our research was to exploit a dynamic network of receptor for discover “intrinsic similarity” of a set of molecules. The term “intrinsic similarity” means that the dynamic network itself will set his own parameters for assessing similarity. 


How it works? And what is a dynamic network?

In chemistry some reactions are irreversible, some are reversible. A dynamic network is formed by linking different chemical building blocks through reversible reactions. This means that the building blocks can exchange with one another for forming different products. 


In this paper three different building blocks are able to form six different macrocycles. The amount of each macrocycle is variable and it’s dependent by the addition of an “effector”. For example if we add an effector that can be bound to one specific macrocycle, the amount of this latter will increase. This also means that we can use the concentration of each single macrocycle as variable on the addition of different effectors. We can simply add a number of different effectors and then compare the six variables of the dynamic network after each addition. If two or more effectors change those six variables in a similar way, this means that those effectors are similar. In a few words, we are using this dynamic network for generating a number of receptor for different effectors.


Comparing six variables is straightforward and it can be done by hand. However a computer is way faster then us in comparing different numbers. We used a basic algorithm for comparing all the six variables (clustering analysis). 

As preliminary test we used 25 different effectors. The selection was done based on chemical groups that are already used as drugs or have something in common with well known drugs. This specific dynamic network was able to distinguish two big sets of molecules: one with ethylamine group and one without it. Apparently, the discriminant of this specific dynamic network was the presence of ethylamine moiety. 


We tested also another approach. We forced another algorithm to see the similarity that we want to screen. We teach the computer which molecules are similar for us (for a specific parameter that we want to screen) and train it with the variables. After that the algorithm was trained to screen for “our” similarity, we fed it with some unknown molecules. At this point it compared this variables with the ones acquired during the training and it told us if this unknown was more similar to a set or molecules or to another.


In principle this can be used for screening for biological active molecules. We can train the algorithm with molecules that are active on a specific protein, or are liposoluble, or are DNA intercalator and so on. Then we can simply test a new synthesized molecule with the dynamic network, and the algorithm will put this new molecule in one of the classes.



This specific dynamic network was used only to proof the concept of assessing molecular similarity using a dynamic network. Now it is time to use more complex networks. More and more different macrocycles mean more receptors, more variables and possibly a better discrimination.

Another great point of this approach is that it can be completely automated. You can use robots for mixing the samples, injecting in the HPLC for the analysis, another computer could read the output and do the analysis. All of this without your presence:


Then again, if you want to read the paper, it’s “free” and open access on the Journal of Systems Chemistry


3 Thoughts on “Paper Explained: Using thermodynamically controlled networks to assess molecular similarity

  1. Really nice read, explaining papers has been something I’ve considered doing for a while, but never found the right paper to start with. I was just curious what kind of level (of chemistry-education) you were aiming for when writing it?

    • I hope to reach non-chemists. But for understanding this one you probably need a little bit of chemistry. I’m looking for feedback on so many levels, but it was important to start. I chose this paper because, well, it’s mine, and it’s open access. So whoever is interested in the paper can download and read it.
      I think it’s a good idea, and if you are planning to do it, just do it. Pick up a paper that you like, or it’s interesting, and explain it without too much chemistry.
      cheers and thanks

  2. Very nice! I love this kind of stuff. If you’re into materials science I know MRS has a website ( with a lot of explanatory articles.

Post Navigation