Researchers develop AI-based program that speeds up discovery of new polymers

Image Credit: Adobe Stock/See Less

It’s no exaggeration to say that the functionality and popularity of artificial intelligence (AI) are growing by the day, as the technology itself improves. AI is the ability of a system or a program to think and learn from experience, and applications that use it have advanced tremendously in recent years, and are now being used in practically every business industry.

Polymer development, it turns out, is no exception, as a team of interdisciplinary researchers from the Faculty of Natural, Mathematical and Engineering Sciences at King’s College in London, UK, has just demonstrated. The team has developed a new software program called PySoftK that uses AI to identify new polymer materials, which could be used across a wide range of applications including in medical technology, pharmaceuticals, energy storage, and more.

Polymers, as almost anyone in plastics knows, are large molecules made up of smaller repeating molecules called monomers, which bond together in a chain-like fashion to form a long polymer molecule. Polymers can be artificial – as in plastics and synthetic fibres – or natural. Using computers to develop polymers isn’t new – for decades now, computer simulations representing 3D structures of molecules have been used to improve the understanding of the relationship between chemical structure and function in increasingly complex polymers; and recent advances in computing power and computational algorithms have enabled scientists to investigate more complex systems and provide more accurate predictions using molecular-scale simulations at speed, leading to faster and more cost-effective design of materials.

But the new PySoftK software takes this a step further, the King’s College scientists say, by providing a robust dataset for researchers to train AI to identify desirable polymer properties.

“PySoftK will allow us to accelerate the development of novel polymers for a whole range of applications, from using polymers with embedded nanoparticles to stitch human tissue together, to improving energy storage methods,” said Professor Chris Lorenz from the Department of Physics and a lead researcher on the technology. “These materials will help form a building block to tackle large scale challenges that we face in health care, in developing biodegradable home and personal care products and in creating more environmentally friendly energy storage systems.”

Normally, Lorenz said, maintaining a large, diverse and accurate molecular database is a hugely costly and time-intensive process, as researchers race to label and categorize models correctly. “By offering a set of tools and programming modules to automate the process of curating, modelling and creating libraries of polymers, PySoftK facilitates the generation of large databases on which to train future machine learning [ML] and deep learning [DL] models,” Lorenz said. “This allows researchers to move their focus away from exhaustive library maintenance and onto discovering new materials.”

The PySoftK software package is described by the King’s College team as versatile, flexible, and easy to install; and it can generate a wide range of polymer topologies and perform library generation in a fully parallelised manner, which makes it highly efficient. The researchers hope that these models will be the driving force of new designer polymer development. PySoftK could also play a significant role for researchers for new functional materials in nano- and bio-technology, they said, since without reliable data to train the AI, they risk making inaccurate predictions.

Source: King’s College London