New Cleansing Agent Development Process Utilizing AI And Molecular Simulation Technologies Examined: Collaboration with Kirin Holdings, FANCL, and Keio University

Research Background and Objectives

FANCL has conducted a variety of research on cleansing agents. In recent years, in response to the need for shorter work hours as women enter the workforce and to the growing awareness in society of the need to reduce environmental impact, FANCL has begun development of a water-based cleansing agent that combines the functions of a cleansing agent and face wash in a single product. By employing machine learning and molecular simulation technologies in the product development process, FANCL conducted joint research with Kirin Holdings and Keio University with the aim of rapidly finding the optimal combination of cosmetic ingredients with high makeup removal performance out of an infinite number of possible combinations and bringing an innovative product to market.

Research Methods and Results

Since 2016, FANCL has calculated the removal rate of makeup stains with commercially available waterproof eyeliner and artificial leather, and has accumulated more than 500 actual measurement data by calculating the cleansing rate. This time, this data was utilized with Kirin Holdings’ machine learning technology to create a system that predicts the performance of makeup remover.

In addition to data on the composition of the cleansing agent components and cleansing rate, information obtained from the molecular structure of the ingredients used - such as chain length and molecular size - was quantified using a database of ingredients and added to the machine learning technology to predict the cleansing rate with high accuracy (coefficient of determination*3 R2 = 0.765, see Figure 1).
*3 An indicator of the goodness of fit of values estimated by machine learning techniques, taking values from 0 to 1, with values closer to 1 indicating that the predicted value fits the actual data.

Kirin Holdings' AI Machine Learning Technology

Kirin Holdings has been accumulating knowledge of machine learning technology through the development of Brewing Takumi AI machine-learning beer development support system and other methods that support the development of new beer products in order to dramatically improve operational efficiency and create value in the time created. There are more than 10,000 types of cosmetic ingredients, and the number of combinations of these ingredients is enormous. By utilizing Kirin Holdings’ machine learning technology in the development of cleansing agents, we aim to develop highly functional products in an efficient manner.

Cleansing Rates Predicted and the Cleansing Rates Generally Matched

Using the cleansing rate prediction system constructed in this study, 100,000 calculations were performed on a computer using Kirin Holdings’ machine learning technology. The results showed that among surfactants, FANCL's proprietary ingredient, eicosaglycerol hexacaprylate*4 (EH), had the highest makeup removal performance, and its combination with certain ingredients such as PPG-9 diglyceryl ether*5 and cyclohexylglycerin*6, which are moisturizing agents, tended to contribute to high cleansing rates (Figure 2).

From the results of 100,000 calculations, a prototype cleansing agent with the one of the highest predicted cleansing rates containing the aforementioned ingredients was created and evaluated for makeup removal performance. The predicted and actual cleansing rates generally matched. As shown in Figure 3, the model cleansing agent was confirmed to have high cleansing performance not only for various types of makeup, but also for various types of dirt such as sebum and mud.
*4 A cleansing ingredient classified as a nonionic surfactant. It is a raw material originally developed by FANCL and has been used in products such as mild cleansing oil.
*5 Moisturizing ingredient that blends well with the skin with little stickiness despite high moisturizing effect.
*6 Moisturizing ingredient that blends well with the skin with little stickiness despite high moisturizing effect.

Molecular Simulation Technology Reveals the Mechanism of Cleansing Agents

Keio University's molecular simulation technology was incorporated to further improve prediction accuracy. The makeup removal performance of a cleansing agent is considered to be strongly influenced by the structure of the surfactant molecules used, as well as the aggregate structure forming the surfactant and its adsorption state on dirt. Keio University used a molecular simulation technique called the DPD method*7 to simulate the structure and behavior of cleansing agent mixtures on a computer. After simulating more than 100 prescriptions, a correlation between the actual cleansing rate and the simulation results of how dirt is absorbed was found.

Furthermore, it became clear that there are characteristic differences in the structures formed in water between cleansing agents with high and low cleansing rates (Figure 4). In the future, we will incorporate the data obtained from the molecular simulation results into machine learning to improve the accuracy of the cleansing rate prediction system and to study its practical application.
*7 A type of molecular simulation, an analytical method that can reproduce phenomena in a more multicomponent and complex mixture of systems than conventional molecular simulations by treating atoms and molecules as a single group to some extent.