Efforts to develop an alternative to animal testing for evaluating eye irritation using machine learning (Update), progressing toward a more accurate predictive model
- Joint research between Nagoya City University and Rohto Pharmaceutical
December 24, 2025
ROHTO Pharmaceutical Co., Ltd. (Headquarters: Osaka City; President: Hidetoshi Seki) is developing cosmetics (including medicated cosmetics) without animal testing. In collaboration with the Abe Laboratory of the Graduate School of Data Science and the Department of Drug Safety Assessment of the Graduate School of Pharmaceutical Sciences at Nagoya City University (Location: Nagoya City; President: Kiyofumi Asai), we have been developing an in silico *1 predictive model for evaluating the eye irritation potential of poorly soluble substances. As a result, we have successfully constructed a new model that can predict eye irritation potential based solely on chemical structure information. Furthermore, we have now developed a new model that enables more detailed classification of eye irritation potential. The results of this research were presented at a symposium and poster presentation at the 38th Annual Meeting of the Japanese Society for Alternatives to Animal Experiments (held November 1-3, 2025).
Key points of the research
- Using in vitro test data from eye irritation tests (STE test method * 4), we have developed a new in silico model process that predicts the eye irritation potential of chemical substances solely from chemical structure information.
- Confirmed the effectiveness of a new highly accurate prediction model that combines a Category 1 *2 prediction model and a non-Category *3 prediction model
- We have discovered the possibility of predicting and screening the toxicity of many chemical substances to the eyes in advance, including not only poorly soluble substances but also chemicals that cannot be synthesized.
Research Background
In the evaluation of eye irritation, in vitro alternative test methods based on the Organization for Economic Cooperation and Development (OECD) test guidelines have shown a certain degree of effectiveness, but there is an issue in that they cannot be applied to some chemical substances, such as poorly soluble substances.
To address this issue, in silico models using machine learning *6 are gaining international attention as a method for predicting toxicity by utilizing existing data and chemical structure information.
Until now, we have been developing a model to predict eye irritation from structural information and physical properties of chemical substances using in vitro test data from the STE test method, which has been established as a guideline as an alternative to animal testing in the Draize test*7. However, we have now begun work on building a new model that can predict both Category 1 and Out of Category with high accuracy.
result
Based on the consistency with the toxicity classification (GHS classification *8) obtained from the STE test, in silico models were created for the Category 1 model (5% and 0.05% concentration solutions) and the Non-Category model (5% and 0.05% concentration solutions) using machine learning with a gradient boosting decision tree algorithm, and their performance was evaluated.The evaluation results were nearly identical between in vivo *9, in vitro, and in silico, and a highly accurate prediction model was successfully developed.
Impact of this research result on society (significance of this research result)
This model is capable of predicting STE tests using only chemical structure information of chemical substances.
This system is expected to be used not only to confirm the safety of poorly soluble substances and chemical substances that are difficult to synthesize, but also to select candidate ophthalmic ingredients and evaluate the dangers of cosmetics in the event of misuse.
Rohto Pharmaceutical's Initiatives to Alternatives to Animal Testing
In addition to the above efforts, we are also promoting research and development activities that proactively utilize in silico toxicity prediction, which involves simulations and data analysis using computers and information technology, in vitro evaluation methods using cultured cells, and in chemico *10 evaluation methods, which use analytical equipment to evaluate the reactivity of biological substances with chemical substances.
In order to continue providing products and services that contribute to the well-being of our customers, we will promote our efforts in research into alternatives to animal testing.
Terminology
*1: In silico
Research methods such as simulation and data analysis using computers and information technology.
*2: Category 1
Test chemicals that cause serious damage. UN GHS Category 1.
*3: Not classified
Test chemicals that do not require classification for serious damage. UN GHS Not Classified.
*4: STE test method (Short Time Exposure Test)
This is an alternative method included in the test guidelines of the Organization for Economic Cooperation and Development (OECD). It is an international test method that enables GHS classification of test substance solutions as non-irritants or strong irritants based on cell viability after exposing corneal epithelial cells to 5% and 0.05% concentrations of the test substance solution for 5 minutes.
*5: In vitro
Tests conducted in artificial environments such as test tubes or incubators.
*6: Machine learning
Computers (machines) analyze (learn) large amounts of data to find patterns and regularities. This is a technology that improves the accuracy of predictions and decision-making.
*7: Draize test
An eye irritation test primarily using rabbits, listed in the test guidelines of the Organization for Economic Cooperation and Development (OECD).
*8: GHS classification
The Globally Harmonized System of Classification and Labelling of Chemicals (GHS) was adopted as a United Nations recommendation in July 2003. GHS classifies the hazards of chemicals according to certain globally unified standards, with the aim of helping to prevent accidents and protect human health and the environment.
*9: In vivo
A test that evaluates in vivo reactions using animals, etc.
*10: in chemico
A simple and quick research method that does not use living tissue or cultured cells. It is a test method that mainly evaluates only the chemical reaction of a substance.