In what could be considered a boon for pharma and other chemistry-driven industries, researchers from the Indian Institutes of Technology in Delhi and Madras have developed a machine learning tool that can fish out new ways of synthesising novel molecules.
“Suppose you want to synthesise a new molecule and you know its structure from other studies, but not its potential method of synthesis, our tool lets you know whether an organism can help you make it or not,” said IIT Madras biotechnologist Karthik Raman.
The algorithm that Raman and IIT Delhi computer scientist Sayan Ranu created does this by mining the repertoire of biochemical reactions available in public databases maintained by research groups across the world. The scientists who recently reported their work in the journal Bioinformatics, have aptly named the software tool Reactionminer.
Aravind Sankar, a dual degree student in IIT Madras' Computer Science Department, also contributed to the work.
“Given that so much information about different types of biochemical reactions in a variety of organisms is already available in the public domain, the knowledge captured in these repositories can be exploited to predict how a new compound can be made,” said Raman, who is also with the Bhupat and Jyoti Mehta School of Biosciences at IIT Madras.
The tool would be particularly useful for chemical and pharma companies as well as for biologists who would like to explore why a biochemical reaction happens in a particular way when dozens of different chemical routes exist, said Ranu, who was with IIT Madras till December last year.
“The system that we developed offers (bio)chemists a way to play around and explore different options available to them. Typically, all such predictions are currently done by humans who depend heavily on what they have learned right from their school chemistry classes and go on to base themselves on intuition,” said Ranu.
“With this tool, they can do it on a scale. They do not have to limit themselves to just one option or two. It is capable of giving them a number of options,” he said.
“This is what machines are good at,” Ranu explained.
To develop this software tool, which “learns” from knowledge available in public databases, the IIT researchers used Graph theory, a mathematical theory which in simple terms breaks down information into points, lines nodes and is commonly used by social scientists like sociologists.
The scientists used this to understand how chemical bonds are broken and formed to make new molecules and how enzymes act upon different classes of molecules.
Explaining it further, they said it was somewhat like Google Maps. When someone has to move from location A to location B, he or she is presented with a number of routes. But based on the mathematical modelling, Google Maps offers the person the fastest possible route.
And this happens inside our bodies too. For example, when we take food, it gets converted into sugar through a process called glycolysis. A particular chain of reactions occurs when the food is converted into glucose. Interestingly, this pathway is followed in most organisms, however big or small they are.
“This doesn’t mean there are no other possible pathways available. There are, in fact, hundreds of them. But surprisingly, it is conserved in most species,” said Raman.
Elaborating on its potential applications, Raman said it could find application in industries where metabolic engineering – where the metabolic machinery of a living organism is exploited to make molecules of import – is used. Currently, a range of chemical industries use bacteria like E. coli or fungi like yeast to synthesise industrially important molecules such as biopolymer or antibiotics.
“In 2006, there was a famous work in which scientists took the anti-malaria drug artemisinin and almost synthesised it in an organism – they stopped shy of just one or two final steps, which were carried out in chemical reactors,” he said.
As a next step, the IIT scientists plan to host an open web server based on their algorithm that could be used by anyone who is interested.
