Protein assemblies in the human body can lead to the development of many diseases in the brain or the blood. This is why researchers need to be able to analyse the assemblies to understand the diseases and eventually cure them. This analysis used to take weeks, but now artificial intelligence can do it in minutes.
Proteins do not always behave as desired in the human body, and this can lead to the development of numerous diseases.
For example, Alzheimer’s disease results from protein assemblies in the brain, leading to the formation of harmful amyloid plaques.
Traditionally, researchers have analysed high-resolution images of protein assemblies, often manually, to dive deeper into understanding Alzheimer’s disease or possible treatments. This often takes weeks, since each image must be studied closely to count and describe the assemblies.
Ultimately, this process would be easier if artificial intelligence could be used to analyse the data in the images within minutes and tell the researchers the exact content of the sample under the microscope – and researchers at the Department of Chemistry at the University of Copenhagen have just developed a tool for this.
“Analysing protein structures in images taken with high-resolution microscopy has been a bottleneck when investigating many diseases relating to protein aggregation and several other research fields. We are now removing this bottleneck with a solution that can reduce weeks or months of analysis to minutes,” explains a researcher that co-lead the development of the analytical tool, Jacob Kæstel-Hansen, Postdoctoral Fellow, Department of Chemistry, University of Copenhagen, Denmark.
The new tool, which has just been described in Nature Communications, is called SEMORE (Segmentation and Morphological Fingerprinting) and has been developed in the group of Nikos Hatzakis by Steen W.B. Bender, Marcus Winther Dreisler, Min Zhang, Jacob Kæstel-Hansen, and Nikos Hatzakis at the Center for Optimized Oligo Escape and Control of Disease, University of Copenhagen.
Analysing microscope images takes time
SEMORE can be used for analysing protein aggregation and assembly in many different ways, including research on neurodegenerative diseases such as Alzheimer’s or Parkinson’s.
Both diseases are characterised by protein assemblies in the brain, and this has been the focus of intensive research for many years with the aim of preventing these assemblies from forming and thereby eliminating the diseases.
For example, researchers investigating the development of Alzheimer’s in mice repeatedly take brain samples from the mice and assess the development of aggregates over time.
Researchers may want to determine how many protein aggregates are present in a given section of the brain, their shape and size and how they develop over days, weeks or months.
They may also want to determine what happens to protein aggregates if the mice are exposed to experimental drugs.
This thorough research can involve hundreds or thousands of high-resolution images with researchers manually counting and describing thousands of protein assemblies – which takes a ridiculous amount of time.
“Researchers spend way too much time analysing microscopy images of protein assemblies, which delays the research. There has therefore been a great need for automated methods that can reliably determine what is in the images captured under the microscope,” says Jacob Kæstel-Hansen.
Analysing every tiny detail in the images or videos
Based on machine learning, the SEMORE model developed by the researchers can extract all the data the researchers want from the microscopy images.
SEMORE is designed to recognise protein assemblies and can tell within a few minutes how many assemblies an image has and their size and shape.
If SEMORE is fed data from a video file, it analyses every frame in the film and determines how things have developed over time. Have there been more assemblies or fewer? Have they grown or shrunk? Do they change shape over time?
All of this is crucial knowledge for researchers who may want to know how neurodegenerative diseases develop.
“SEMORE is not limited to examining images of protein aggregates in neurodegenerative diseases and can be used in many biological systems for analysing protein assemblies. We performed an experiment in which insulin aggregated, reducing its effectiveness, and we analysed this in very few minutes,” says Steen W.B. Bender, Department of Chemistry, University of Copenhagen.
Creating a library from the data
SEMORE has the potential to ease the work for thousands of researchers who today painstakingly count protein assemblies on static microscopy images, and for them, it is welcome news that the model is both freely available and plug and play.
This means that researchers can analyse their data with SEMORE and rapidly get a file with information that would otherwise have taken them weeks to prepare.
Professor Nikos Hatzakis, who led the work together with Jacob, hopes that as more researchers use SEMORE, data from the many analyses will be collected in a database that can become a research resource.
“For example, we are examining insulin, and if other researchers do the same or study other proteins, we can create a whole library of the aggregation and assembly of various proteins under different conditions. In the long term, this can make us more aware of how various conditions affect disease development,” he concludes.
