What links ML, lasers, and tiny gold-plated micro-bots? Answer: Smart medicines

'Drug delivery would certainly be an intriguing application,' prof tells us

Tiny so-called microswimmers coated with gold can be moved around in a liquid using a laser system controlled by a machine-learning system – and scientists hope the technique will be used in some way in the future to transport drugs inside humans.

Some experts believe nanomedicine, an area of research involving tiny devices capable of performing rudimentary tasks including sensing and swimming, holds much promise for things like drug delivery and surgery. One of the biggest hurdles, however, is controlling these robots once they’re in a petri dish or the human body.

It’s difficult to control such small gizmos in liquid since they’re pushed in random directions by the jostling of other particles, an effect known as Brownian motion. They often have to be manually controlled in the lab using magnets or pulses of electricity. Now a team of researchers led by Leipzig University in Germany, however, believe the whole process can be automated using lasers.

In an experiment described in a paper published in Science Robotics, a microswimmer some 2.18 micrometres across and coated with 8nm of gold was maneuvered in a tiny drop of solution. The microswimmer could be propelled in eight directions – such as up, down, left, right, and diagonally – using laser zaps.


Top doctors slam Google for not backing up incredible claims of super-human cancer-spotting AI


Heat from the laser is absorbed by the gold, and one side of the microswimmer becomes hotter than the other. The difference in temperature causes a pressure imbalance, and propels the swimmer in a particular direction. Depending on the direction of incoming laser pulse, the particle can be nudged one way or another. The researchers trained a reinforcement-learning algorithm to learn how to direct the swimmer, using laser bursts, to a desired location.

“The training is the motion of the particle in the sample with the corresponding actions, that are controlled by the laser,” Professor Frank Cichos, co-author of the paper and head of the molecular photons group at Leipzig University, told The Register.

First in code, then in real life

The motions of the microswimmer were first modeled in software; it’s virtually directed to swim to a specific region and rewarded when it moves closer to the target and penalized when it moves further away from it. Over time, it learns how to swim against the grain of the Brownian motion to reach a particular location as efficiently as possible.

To run the algorithm on a real microswimmer, the motion of the gold-plated particle has to be tracked using a microscope and fed into the software. The reinforcement-learning algorithm then predicts how the device should move and this output is used to control the direction of the laser to help it swim.

This is all happening in real-time, so the particle is exploring the sample, and a microscope and the computer ‘sees’ the particle’s motion and decides what the particle should do

“The algorithms are moving the laser to control the particle using reinforcement learning [techniques running] on the computer,” Prof Cichos said.

"This is all happening in real-time, so the particle is exploring the sample, and a microscope and the computer ‘sees’ the particle’s motion and decides what the particle should do, according to the reinforcement learning algorithm."

This work is at the proof-of-concept level; it takes hours to train the software to move the microswimmer across distances of a few micrometres. Applying it to move a tiny nanobot in blood in a vein or an organ also introduces new challenges. Not only does the laser have to be able to, ideally harmlessly, penetrate skin to reach the tiny particle, the microswimmers have to contend with other things like blood pressure.

“Drug delivery would certainly be an intriguing application," the professor told us.

"For that purpose, it would even be nice, if we could equip such active particles with some intelligence themselves, such that we do not need external control. This is certainly one of the dreams scientists have in this area."

“I think there is still a long way to go. Scientists are exploring the development of active particles for the use inside the body but also for environmental applications. The next steps would require to add some type of external or internal control for example. To guide them with adaptive algorithms that we have used, one needs control over the propulsion mechanisms together with some imaging technique. Light is certainly great here for remote control, but has its limitations when you need to go deep into the body." ®

Biting the hand that feeds IT © 1998–2021