Unlocking Protein Aggregation Through Machine Learning: An Interview with Prof. Nikos Hatzakis

More than a century has passed since pathological protein “packages” were first identified in the brains of patients with neurodegenerative diseases. Many of these diseases have been linked to proteins failing to fold properly or maintain their native structure, rendering them non-functional and prone to aggregation in various brain regions. Scientists worldwide are seeking new treatments for aggregation diseases such as Alzheimer’s, Parkinson’s, and cancer.

The reasons behind protein aggregation and how to treat it remain scientific mysteries. Until now, studying this phenomenon has been challenging due to the lack of appropriate tools. Researchers from Hatzakis Lab at the Department of Chemistry and Nano-science, University of Copenhagen, have developed a machine learning algorithm that significantly aids neuroscientists. This algorithm tracks protein aggregation in real-time under the microscope, enhancing our understanding and treatment of neurodegenerative diseases and cancer.

The algorithm automatically maps and tracks the key features of accumulated biomolecules that cause Alzheimer’s and other neurodegenerative disorders, a task previously impossible. “Within minutes, our algorithm solves a problem that researchers would take several weeks to solve. By studying microscopic images of aggregated proteins more easily, we hope to advance the understanding of neurodegenerative disorders and eventually lead to new treatments,” says Nikos Hatzakis, a Professor at the University of Copenhagen and visiting professor at Harvard Medical School, who led the research team along with Dr. Jacob Kæstel-Hansen.

Detection of tiny proteins and other biomolecules interact and exchange compounds and signals billions of times inside our cells, allowing our bodies to function. However, when errors occur, proteins can aggregate, disrupting their intended function and leading to neurodegenerative disorders and cancer.

The researchers’ machine learning algorithm, called SEMORE, published in Nature Communications, can identify clusters of proteins down to a billionth of a meter in microscopy images. It measures and groups these clusters according to their shape and size, tracking their evolution over time. The algorithm captures structural variations and morphological diversity in different or even the same types of structures. The appearance of biomolecule clusters can significantly impact their function and behavior in the organism.

“Protein aggregates appear under the microscope in various structures, such as filaments or spherical shapes, varying according to the disorder they cause. Measuring these thousands of structures by hand is time-consuming, if not impossible. If we can see, track, quantify, and describe these aggregates and their evolution, we achieve a complete understanding. No other method can do this so automatically and efficiently,” adds Professor Hatzakis.

An open tool for everyone interested in aggregation

Researchers at the Department of Chemistry are now using the algorithm to conduct experiments with insulin molecules. Professor Hatzakis’ team has developed a method to observe how insulin molecules are placed one by one inside cells, affecting their function and impact on blood glucose levels both short and long term.

“Our new tool allows us to see how insulin aggregates respond to added compounds. This model can help us understand how to potentially stop them or make them less dangerous or more stable,” explains Professor Hatzakis.

The team sees great potential in using the algorithm to develop new drugs once the tiny building blocks are clearly identified. “As other researchers worldwide start using our algorithm, we believe a large library of molecular and protein structures related to various disorders and biology will be created over time. This will lead to a better understanding and management of diseases,” concludes Professor Hatzakis.

The SEMORE algorithm is freely available online as open source for researchers and anyone working on understanding the aggregation of proteins and other molecules. This research was funded by the Novo Nordisk Foundation Center for Optimized Oligo Escape and control of disease, led by Professor Hatzakis.

Twin4Promis was honored to welcome Professor Hatzakis as an invited speaker at the 1st Twin4Promis training workshop. He shared his expertise on “Revealing Hidden Patterns in Protein Dynamics and Aggregation through Single-Molecule Studies and Machine Learning”. His profound knowledge and insights into protein misfolding are invaluable to the field.

Read the full article in Greek, here.

Share this post: