Using machine learning, researchers have developed an algorithm that can track the activity of all the particles darting around inside our cells – from lipids and proteins to viruses, medicines and the cell’s own organelles. This technique enables scientists to tackle questions that were previously out of reach.
For years, scientists have dreamed of mapping how particles move inside cells – and how that motion changes over time.
Such insights could reveal how cells work – and uncover the earliest signs of disease, long before symptoms appear. But the challenge is daunting: thousands of types of molecules exist inside cells, each present in the thousands. And each molecule darts about in its own complex, seemingly chaotic path.
Under the microscope, thousands of dots swirl in a dizzying blur – without order or direction.
Machine learning untangles the mess
The problem, however, can be solved with machine learning – and that is exactly what researchers at the Department of Chemistry at the University of Copenhagen and Harvard Medical School, led by Nikos Hatzakis, have done.
More specifically, the researchers built an algorithm that tracks the activity and movement of every molecule in a cell, based on video footage.
“Our method enables cellular insights that would not be possible otherwise. And because the process is automated, you can just feed your data into the algorithm – and it will tell you how specific molecules have moved,” explains first author Jacob Kæstel-Hansen, postdoctoral researcher at the Massachusetts Institute of Technology, Cambridge, MA, USA.
The research was carried out as part of Jacob Kæstel-Hansen’s PhD project at the Department of Chemistry and the Center for Optimized Oligo Escape and Control of Disease under the Department of Chemistry, University of Copenhagen, Denmark.
Inside the cell, nothing moves in a straight line
The interest in studying particle movement inside cells is not only vital but also wide-ranging, says Nikos Hatzakis.
For instance, tracking how viruses move inside cells can help researchers to better understand infections and how they progress. That could lead to new ways of detecting, understanding or combatting disease.
Mapping the activity of molecules such as insulin or other drugs inside cells can also offer insight into how to optimise drug delivery, how to maximise therapeutic effect or how to reduce side-effects.
And tracking the movement of biomolecules and organelles can reveal fundamental details about how cells function when everything is working properly.
“But the problem is that no particle moves in a straight line from A to B. They move chaotically, change direction, swirl around and are influenced by the motion of other particles. Under a microscope, it looks like pure anarchy – like trying to follow thousands of ants scattering in every direction at once. If you tried to track all of that by hand, it would take weeks – and even then, getting a clear overview would be nearly impossible,” explains Jacob Kæstel-Hansen.
The algorithm analyses 40 traits per particle
To make sense of the chaos, Jacob Kæstel-Hansen and colleagues built a machine-learning algorithm, which is like training a computer to recognise faces – but with particles instead of people.
The algorithm can be fed video recordings of particles captured under a fluorescence microscope and will then analyse and describe the activity of each moving particle individually.
It categorises movement over time and assigns up to 40 distinct traits to each particle – such as direction, speed, movement pattern, the area in which the particle moves and much more.
Together, these 40 traits form a kind of digital fingerprint for each particle – revealing what it is doing, how it moves and where in the cell it operates. The algorithm does not just track where a particle is but also what it seems to be doing – whether it is searching for something, transporting something or just drifting.
“The idea is that motion reveals underlying information. It is like walking down the street and observing people. If someone is walking, you might assume that they are not in a hurry – but if they are running, you think something else is going on. This is the kind of information the algorithm can extract,” says Jacob Kæstel-Hansen.
A tool for the whole scientific community
The researchers hope that the algorithm will catch on – and that scientists will share their data to help to build a global knowledge base about the inner workings of cells.
Over time, this could create a massive repository of information on how particles act inside cells.
A shared database could show how drugs travel inside cells – and how healthy and diseased cells respond differently, with or without treatment. Eventually, the algorithm could even help to accelerate the development of new treatments by instantly showing how a drug acts inside cells – molecule by molecule.
“This could provide a much clearer view of what separates healthy from diseased biology – at the level of the cell,” says Nikos Hatzakis, who has also founded EDGE Biotechnologies, a company using this type of information to improve drug delivery.
This is why the team has made the algorithm freely available – to help everyone uncover new insights in their own data. It is available in two versions: one with a simple user interface and one for users with coding experience.
Researchers around the world have already begun using the algorithm.
“The collaborators with whom we have worked tell us that the algorithm has sped up their analysis – and opened the door to questions and insights they could not previously access,” says Jacob Kæstel-Hansen.
This is the first time researchers have seen the cell’s hidden swarm in such detail – and it might just reshape how we understand life itself.
