The creation of this robot began three years ago, but it couldn’t have come at a better time. It began when IBM scientists started developing machine learning models to predict chemical reactions. A few months later, they launched the RXN for Chemistry service, a method of translating this learning based on neural networks that can give the most likely result of a chemical reaction. Today, it is 90% accurate.
The new system utilizes deep learning algorithms, cloud services, and robotic laboratories to automate the entire process and aid the chemists in their work without the need for physical presence at a research laboratory.
RoboRXN is one example of how technology is making progress in the medical industry, incorporating advances in artificial intelligence (AI) and cloud usage to provide solutions more effectively, cheaply, and quickly.
At the right time
Development of the robot comes at a crucial moment for progress in the field of health. The Covid-19 pandemic has set the world off on a race to find a vaccine that prevents contagion. Manual methods could take researchers years, which would mean more lives lost.
On average, it takes about 10 years to discover and commercialize a new product. What’s more, production costs are estimated to be above 10 million dollars.
“We began collaborating with a group working on synthetic organic chemicals from the University of Pisa in Italy to integrate the architecture. All we needed was a combination of AI, cloud technology, and automation of the chemistry. Now, the production of molecules is executed in a remotely accessible laboratory with the least amount of human interaction possible,” said Teodoro Laino, technical leader for molecular simulation.
It’s not yet clear whether this effort will help facilitate development of a coronavirus vaccine, but it will lay the foundation for the next generation of research tools for pharmaceuticals and chemical products and it will guarantee that we are more prepared in future.
RoboRXN also allows molecules to be created directly from laboratory notes, as its AI model has been trained on approximately one million patents. All it needs to work is to copy and paste the paragraphs into a software text box and it will begin the automation process.
“Once the machine learning algorithm has acquired sufficient examples, it will be capable of finding out for itself which words to pay attention to in order to extract the correct production steps. In order to provide the training data for the machine learning model, we established an annotation framework that allowed us to generate examples of phrases related to synthesis procedures and the corresponding operations. The main advantage of this approach is that it is solely based on data. To improve, it will simply need more examples,” added Laino.