Machine learning is a data analysis technique that automates the creation of analytical models. It is a subfield of artificial intelligence that is based on the idea that systems can learn from data, identify patterns, and make decisions with little or no human intervention. Researchers have developed a new approach to machine learning that ‘learns how to learn’ and outperforms existing machine learning methods for drug design, potentially speeding up the search for new disease treatments.
The method, known as transformational machine learning (TML), was created by a group of researchers from the United Kingdom, Sweden, India, and the Netherlands. It learns from a variety of problems and improves performance as it does so. TML could speed up the discovery and production of new drugs by improving the machine learning systems used to identify them. The findings have been published in the Proceedings of the National Academy of Sciences.
The TML method improves performance while learning and can learn from multiple problems at the same time. It could also speed up the discovery and production of new drugs by improving the machine learning systems used to detect them. Most types of machine learning (ML) use labeled examples, which are almost always represented in the computer using intrinsic features such as an object’s color or shape. The computer then creates general rules that connect the features to the labels.
Whereas a typical ML system would have to start from scratch when learning to identify a new type of animal, such as a kitten, TML can use similarities to existing animals: kittens are cute like rabbits but lack long ears like rabbits and donkeys. As a result, TML is a far more powerful approach to machine learning.Professor Ross King
“It’s kind of like teaching a child to identify different animals: this is a rabbit, this is a donkey, and so on,” said Professor Ross King, who led the research at Cambridge’s Department of Chemical Engineering and Biotechnology. “If you teach a machine learning algorithm what a rabbit looks like, it will be able to determine whether or not an animal is a rabbit.” Most machine learning operates in this manner, dealing with problems one at a time.”
However, this is not how human learning works: rather than dealing with a single issue at a time, we improve our ability to learn because we have learned things in the past.
“We used this approach to machine learning to develop TML, and created a system that learns information from previous problems it has encountered in order to better learn new problems,” said King, who is also a Fellow at The Alan Turing Institute. “Whereas a typical ML system would have to start from scratch when learning to identify a new type of animal, such as a kitten, TML can use similarities to existing animals: kittens are cute like rabbits but lack long ears like rabbits and donkeys. As a result, TML is a far more powerful approach to machine learning.”
The researchers demonstrated the feasibility of their concept by applying it to thousands of problems in science and engineering. They claim it holds particular promise in the field of drug discovery, where this approach speeds up the process by checking what other ML models have to say about a specific molecule. A typical ML approach, for example, will look for drug molecules of a specific shape. TML, on the other hand, connects the drugs to other drug discovery problems.
“I was surprised at how well it works, better than anything else we’ve tried for drug design,” King said. “It’s better than humans at selecting drugs, and without the best science, we won’t get the best results.”
Machine learning today is not the same as machine learning in the past due to advances in computing technology. It arose from pattern recognition and the theory that computers can learn without being programmed to perform specific tasks; artificial intelligence researchers wanted to see if computers could learn from data. The iterative aspect of machine learning is important because models can adapt independently as they are exposed to new data. They learn from previous computations in order to produce consistent, repeatable decisions and outcomes. It’s not a new science, but it’s gaining traction.