Chemistry researchers leverage AI for hair dye development by Korean beauty brand Amorepacific

Contact for reporters:
Josh Rhoten
[email protected]

Are you hoping your new hair color will be more hazelnut coffee brown or golden fall harvest? 

Colorado State University chemistry researchers are developing artificial intelligence tools that could enable more targeted, personalized color offerings like those and speed the overall development process for them in the future.  

The project is a collaboration with a Korean beauty manufacturer Amorepacific – one of the largest cosmetics companies in the world – and it’s much more than a simple color-mixing project. The research centers on using machine learning to understand and predict the complex set of chemical interactions required to produce the exact shade. 

“Formulation of hair dye is complicated because a typical product involves many different chemical ingredients to consider, plus their concentrations and interactions with each other. Each instance can subtly change the final color. Our goal is to train AI models that can learn from past experiments and help us predict which combinations are likely to give the color we want,” said Professor Seonah Kim, who is leading the project at CSU. 

As a chemist and computer scientist specializing in AI, she said her team is uniquely positioned to develop the algorithms needed to solve this problem. 

“While we are talking about differences in shades or the search for new colors, this type of complex mixture problem is also an ongoing challenge in the field of chemistry, beyond this example. There is a great need for this rapid chemical and physical property prediction in sustainability work, which my lab also participates in,” Kim said.  

The researchers aim to develop machine learning tools to identify promising patterns in existing datasets and leverage that information to predict new hair dye formulations worthy of further testing, making the search for promising combinations faster and more efficient.  

Kim said the resulting tools from the project will provide the initial steps towards development of highly personalized hair color dyes by helping companies explore a much wider space of possible formulations than would be practical with traditional trial-and-error experiments alone. That’s significant, given that more than half of adult women worldwide have colored their hair, according to industry estimates. 

“It would obviously be difficult for a company to change and run experiments to test each potential formula on its way to providing just the right desired shade of brown a customer is looking for. This machine learning approach can quickly explore potential combinations and highlight which may be most promising to pursue,” she said. 

Kim said the research project started because of a past role with the company she held before joining CSU. Her research group in the Department of Chemistry more broadly uses computational tools to design and optimize chemical reactions that can replace petroleum-derived products with sustainable alternatives. One focus area is developing catalytic strategies to convert biomass, including wood and other plant materials, into high-value renewable chemicals, polymers, plastics, and biofuels. Computational modeling can predict promising reaction pathways and catalyst designs to that end, which are then also validated through laboratory experiments. 

Doctoral student Hojin Jung is leading the machine learning modeling efforts in the research. During his time at CSU, he has led several related projects to develop neural networks that can quickly expand the range of potential chemicals analyzed at a time and reduce long-term prediction errors. He served as the first author on a 2025 paper that used AI tools to predict how well a substance dissolves in complex mixtures of multiple solvents, for example. 

He said his background in chemical engineering lent itself to this research. 

“I am interested in applying the machine learning techniques that are being developed to real-world chemistry problems. The hair dye mixture project is a perfect example of that,” he said. “In earlier projects, we used AI to predict properties like solubility and diffusion in complex chemical systems. Those tools are now helping us tackle mixture problems where understanding how many different chemicals work together is critical.” 

Kim said this type of research could be useful across many industries by reducing reliance on expensive, time-consuming experimental methods. She pointed to the pharmaceutical industry’s search for new drug candidates and efforts in materials sciences for sustainable manufacturing and renewable energy. 

“This project is just one example from our team of how machine learning can be used to enhance predictions, streamline processes, and drive innovation in chemistry,” she said. “Rather than testing thousands of possible formulations, researchers can use the models we develop to rapidly screen candidates and focus experiments on the most promising options.”