Fingerprinting proteins enables researchers to more easily identify interesting industrial or pharmaceutical properties.
Imagine you are producing enzymes for laundry detergent and want to find new enzymes that make clothes even cleaner than is possible today.
This process is hugely expensive because you must carefully review the properties of thousands of proteins to find a few that may be of interest.
However, there is another way: fingerprinting the proteins.
Researchers from the University of Copenhagen recently discovered that two proteins with the same fingerprint also display similar properties. If a protein has a fingerprint similar to that of an enzyme used in laundry detergent, then the manufacturer should carefully assess that protein to determine whether it may actually be a more effective enzyme for detergent applications.
Researchers have now verified the fingerprinting method, which involves fluorescent molecules, a camera to record how the proteins move and then artificial intelligence to make sense of it all.
“Identifying enzymes for laundry detergent is a very expensive process involving many costly experiments. A major advance would be spreading thousands of proteins on a surface and then using their fingerprints to identify the handful of proteins that are most interesting for further investigation,” explains a researcher behind the new study, Henrik Pinholt, a PhD student at Massachusetts Institute of Technology in the United States.
Henrik Pinholt carried out the research in Nikos Hatzaki’s research group at the Department of Chemistry of the University of Copenhagen in collaboration with Wouter Boomsma from the Department of Computer Science.
The research has been published in the Proceedings of the Natural Academy of Sciences of the United States of America.
Tiny luminous markers on proteins
The hypothesis behind the research was that proteins probably have similar properties if they have the same motion.
Henrik Pinholt and colleagues studied the motion of proteins by analysing data recorded with a microscope and a camera. The data are obtained by marking proteins with fluorescent proteins, enabling thousands of proteins to be recorded simultaneously.
By videotaping the proteins, the researchers observe the motion trajectory of numerous luminous markers, and a computer with a machine-learning framework can then deduce which proteins are of interest based on their motion.
The algorithm examined 17 features chosen to describe the behaviour of most of the trajectories.
“The motion trajectory of proteins comprises many features: how fast they move and over what distances and whether they move straight or in a pattern. These many features can be coalesced into something artificial intelligence can recognise. The camera-based technology is the same one that is used for following movement patterns in a football match. This provides a data track that an algorithm can follow,” says Henrik Pinholt.
Motion mimics function
The technology enabled the researchers to identify molecules of interest, and with an accuracy of 90% they can tell which protein is which among the “thousands of proteins in a bucket”, as Henrik Pinholt puts it.
The accuracy exceeds 90% in differentiating two proteins that are already known.
However, the researchers’ primary interest is not identification but function. They want to know not only which molecules are involved but also what they can do.
“We have examined data from similar experiments with drug nanocarriers, and they show that some motion trajectories are linked to the ability to deliver cargo to the body. This indicates which nanoparticles may be pharmaceutically interesting. The same framework can be adopted for proteins,” explains Henrik Pinholt.
Like detecting a shoplifter
According to Nikos Hatzakis, who led the research, imagining shoplifters in a supermarket can help people to understand the potential of the new technology.
Shoplifters may move slightly differently than regular customers because they are more nervous and are constantly checking out their surroundings.
Filming all the shoppers in the supermarket and analysing their movement patterns could enable you to identify the shoplifters as opposed to people buying groceries.
The same principle applies to identifying features in proteins. Researchers can rapidly scan thousands of proteins and determine which proteins have the properties they seek.
“Imagine that you spot some movement patterns in an enzyme in laundry detergent that are linked to its ability to clean clothes. These movement patterns can then be searched for among thousands of proteins at the same time. This can be done in pharmaceutical research for known proteins that exhibit interesting pharmaceutical properties, such as those related to cancer. We would like to discover many proteins that can do something similar. Of course, the technology cannot be used to validate that a protein has a pharmaceutical property, but we can select a handful of candidates among thousands of proteins and then further investigate those that look most interesting,” says Henrik Pinholt, adding:
This is very important for many proteins that are essential for developing drugs, including receptors and proteins that control metabolism or cancer. Understanding the pathways that drugs or viruses use to enter a cell is also important. The Hatzakis Group is actively researching this.