LSMO, based in Sion, is specialized in synthesizing and simulating metal-organic frameworks (MOFs) – a special type of compound discovered some 20 years ago. MOFs consist of metal ions linked by organic molecules to form 3D crystals. Because their molecules can be combined in an almost infinite number of ways, MOFs offer promising opportunities in a wide range of applications. Chemists at LSMO are studying MOFs that can absorb CO₂ in order to develop a system for removing this powerful greenhouse gas from the atmosphere.

The hitch is that developing new MOFs requires a huge amount of time and energy. This kind of chemical synthesis involves optimizing many different variables – solvent composition, temperature, and reaction time, to name just a few. And the more variables there are, the higher the number of possible combinations; researchers can easily find themselves with millions of experiments to carry out to come up with just one MOF. What’s more, the chemical links and assembly processes underlying the formation of MOFs are still not fully understood, meaning there are not yet any basic principles that chemists can follow. They essentially have to start from scratch each time.


Ten billion days

“That’s where intuition comes in,” says Berend Smit, head of the LSMO. “With our research, we wanted to leverage Machine Learning technology to develop a systematic method for quantifying the lessons learned from prior experience.”

His team took as an example an MOF that is well known to scientists: HKUST-1. Its crystalline structure can vary depending on what chemical group is used to synthesize it. To measure how extensively intuition plays a role in synthesizing the right kind of material, the LSMO chemists first used a method that does not rely on intuition at all – a high-performance robotic synthesizer. Their synthesizer processed no less than nine different variables to reverse engineer the process and compile all possible failed synthesis experiments for an HKUST-1 molecule.

“Our robot can crunch through about 30 chemical reactions a day. But even with that high level of throughput, it would still take nearly ten billion days to run through all possible reaction combinations. So researchers working under normal conditions – that is, without a robot – clearly have to rely on intuition to rule out a vast number of possible combinations and focus on the most promising,” says Kyriakos Stylianou, head of chemical synthesis at LSMO.

In other words, whether they realize it or not, researchers who carry out several experiments – successful and otherwise – get a feel for what will work and what won’t. This “gut feeling” tells them which variables could have the greatest influence on the outcome of a chemical reaction. For example, if a scientist finds that changing the reaction temperature alters the results of his experiment, even slightly, then he will be more likely to focus on the temperature variable.


Convincing the scientific community

The Machine Learning method that was developed by LSMO enabled chemists to not only quantify researchers’ intuition, but also program their robot to carry out synthesis reactions more efficiently. That matters because around 1,000 new MOFs are developed each year, and behind each one lie anywhere between 10 and 100 failed attempts. These mistakes contain important and potentially useful information for further research by anyone working in the same field. Thanks to the method developed at LSMO, as well as the platform made available under Switzerland’s NCCR MARVEL program, the lessons learned can be compiled and shared.

“Now we need to convince scientists to open up and talk about their unsuccessful experiments. If we’re able to do that, we can dramatically change the way chemical research is performed,” says Smit.


References

"Capturing chemical intuition in synthesis of metalorganic frameworks", Seyed Mohamad Moosavi, Arunraj Chidambaram, Leopold Talirz, Maciej Haranczyk,
Kyriakos C. Stylianou, Berend Smit. Published in Nature Communications on February 1st, 2019. DOI: 10.1038/s41467-019-08483-9.

By Sarah Perrin. Mediacon. EPFL. Posted:  Feb 01, 2019.

<< voltar para novidades