Machine learning unlocks the superior performance of light-driven organic crystals

Researchers have developed a machine learning workflow to optimize the output power of light-activated organic crystals. Using Lasso’s regression, identifying the substructure of the major molecules and Bayesian optimization for efficient sampling achieved a maximum blocking force of 37.0 mn, making it 73 times more efficient than traditional methods.

These findings published in Digital Discovery will help develop remotely controlled actuators for medical devices and robotics, supporting applications such as minimally invasive surgery and precision drug delivery.

Materials that convert external stimuli into mechanical motion, known as actuators, play a key role in robotics, medical devices, and other advanced applications. Among these, optical mechanical crystals deform in response to light, making them promising for lightweight, remotely controlled operation. Their performance depends on factors such as molecular structure, crystalline properties, and experimental conditions.

A key performance indicator for these materials is their blocking power. This is the maximum force exerted when deformation is completely restricted. However, due to the complex interaction of crystal properties and test conditions, achieving high blocking forces remains difficult. Understanding and optimizing these factors is essential to expanding the potential applications of photomechanical crystals.

In a step to optimize the output power of light-activated organic crystals, researchers at Waseda University are using machine learning technology to improve performance. This research was led by Associate Professor Takuya of the Data Science Center, and Professor Ishizaki Kawakawa and Asano, along with Tagama of the Faculty of Advanced Science and Engineering at the Graduate School of Advanced Science and Engineering at Soseda University.

“We’ve realized that machine learning simplifies searching for optimal molecules and experimental parameters,” says Taniguchi. “This allows data science technology to be integrated with synthetic chemistry, allowing us to quickly identify new molecular design and experimental approaches to achieve high-performance results.”

In this study, the team utilized two machine learning techniques: regression of lasso (minimum absolute contraction and selection operator) for molecular design and Bayesian optimization for selecting experimental conditions. The first step generated a material pool of salicylideneamine derivatives, and in the second pool, efficient sampling from this pool was efficiently sampled for real-world force measurements.

As a result, the team maximized blocking power, achieving up to 3.7 times the power output compared to previously reported values, at least 73 times more efficient than traditional trial and error methods.

“Our research shows a major breakthrough in photographically activated organic crystals by applying machine learning,” says Dr. Taniguchi. “We have demonstrated the potential to dramatically improve the performance of photoresponsive materials by optimizing both molecular structure and experimental conditions.”

The proposed technology has broad implications for remotely controlled actuators, small-scale robotics, medical devices, and energy-efficient systems. Photographed crystals react to light, allowing for contactless and remote operation, making them an ideal robotic component that operates in limited or sensitive environments. The ability to non-invasively generate forces using focus light may also be valuable for microsurgical tools and drug delivery mechanisms that require accurate and remote actuation.

By leveraging clean energy inputs (light illumination), these materials are promising for environmentally friendly manufacturing processes and devices aimed at reducing overall energy consumption, whilst maximizing mechanical output. “Beyond improving force output, our approach paves the way for more refined and miniaturized devices, from wearable technology to aerospace engineering and remote environment monitoring,” adds Dr. Taniguchi.

In conclusion, this study highlights the power of machine learning-driven strategies to accelerate the development of high-performance optically operated materials, bringing it one step closer to practical applications and commercial viability.