For most mortals, chemistry is a distant memory of childhood dating back to our school days, where we were able to experiment with chemical reactions. But chemistry is everywhere and plays an essential role in products and technologies that we probably can’t imagine living without.
However, what most of us may not realize is that, on average, it takes at least 10 years to discover a new material and bring it to market, and the estimated production costs are around $ 10 million. . Research on nylon, for example, began in 1927 and it was first used in a toothbrush in 1938.
Synthetic chemistry or the art of manufacturing materials is still the most traditional discipline in terms of digitization and acquisition of new technologies. Chemists still rely on many of the same protocols from decades ago, and little progress has been made in modernizing old trial-and-error practices to enable a new era of accelerated discovery. And it is at this point that a dynamic group of scientists from IBM Research Europe enter the scene, setting out to change these aforementioned protocols using modern tools such as artificial intelligence (AI), cloud computing technology, and robotics.
IBM scientists
It all started three years ago when IBM Research began developing machine learning models to predict chemical reactions. After a few months of in-house development a free service was launched, via IBM Cloud, which they called RXN for chemistry and which is a next-generation neural machine learning translation method that can predict the most likely outcome of a chemical reaction using neural machine translation architectures. Similar to Italian to English translation, our method translates the language of chemistry by converting reactants and reactants into products, using the SMILE representation to describe chemical entities.
In 2019 scientists from IBM Research Europe began collaborating with a group of synthetic organic chemists at the University of Pisa, Italy, to integrate a retrosynthetic architecture into the RXN tool. To explain it, they thought of making a pizza. The retrosynthetic architecture indicates the ingredients of the pizza and also provides high-level guidelines for creating the pizza in the correct order. Working with the team in Pisa, they added this feature to RXN for Chemistry last October.
How to make chemistry fun again?
The IBM researchers believe the way to do this is to completely rethink chemistry by combining AI, cloud technology, and chemical automation. This mix led to the creation of RoboRXN – machine learning algorithms that autonomously design and execute the production of molecules in a remotely accessible laboratory with as little human intervention as possible.
The main challenge in chemistry is that many operational details, such as “cooking” a pizza, are done in the form of unstructured data, which frustrates easy analysis and interpretation. In order to build an AI model with the ability to learn the correct steps of chemical procedures, IBM scientists designed an algorithm that specifically extracted synthesis information for organic chemistry and converted it into an automated structure and user-friendly format. .
Data-driven schema
Once the machine learning algorithm acquires enough examples, it can figure out for itself which words to pay attention to in order to extract the correct production steps. To provide the training data for the machine learning model, an annotation framework was created that allowed the generation of examples of sentences related to the synthesis procedures and the corresponding operations. The main advantage of this data-driven approach is that it is based on data only. To improve it, you simply need more examples.
Unlike other approaches, the deep learning model converts experimental procedures as a whole into a structured and easy-to-automate format, rather than scanning text for relevant information. Furthermore, it is not based on the identification of individual entities in sentences, nor does it require specifying to which words or groups of words the synthesis actions correspond, which makes the model more flexible and reliable.
RoboRXN learns
Building a reliable field data set for chemical procedures allowed IBM experts to build the core of RoboRXN technology, namely an AI model that, trained in a large number of chemical recipes, learns the details. of chemists to be able to recommend the correct sequence of operations to “cook” a specific target molecule. Going back to the pizza analogy: let’s imagine an artificial intelligence model that can not only retrieve your favorite recipes on demand, but can also automatically extract information from your built-in knowledge to deliver an optimal list of instructions for making that gourmet pizza that you are sure to. will impress your dinner guests.
In IT terms, this amounts to having an artificial intelligence architecture that writes programs to produce molecules or cook food. The scientists’ goal in building RoboRXN was to use this AI model to eliminate the tedious human task of programming commercial automation hardware. And to make the RoboRXN system even more convenient and easy to use, the entire suite of services was deployed on the IBM Cloud to be accessible anywhere there is an internet connection.
What are the implications of this IBM invention? Could an automated system like RoboRXN help chemists cut the discovery period of a new treatment for COVID-19 or any other virus in half? Or what if RoboRXN could help accelerate the development of a fertilizer that does not require consuming between 1% and 2% of the world’s annual energy supply for its production?
