Ask a RIFM Scientist: Why is organizing fragrance materials so important?


Everyone wants safe products without having to test them on animals.

A process called “clustering”—or organizing chemicals into structurally similar groups—has helped scientists at the Research Institute for Fragrance Materials (RIFM) assess thousands of ingredients without having to test them on animals.

Clustering, also called “grouping,” has gained worldwide recognition over the last decade. By organizing materials, scientists have learned how to use data from one material to accurately predict how other similar materials may affect human health and the environment, a process called “read-across.”

In places such as the European Union, which forbid new animal testing for cosmetics and other consumer products, agencies like the European Chemicals Agency (ECHA) encourage and help guide scientists using this innovative approach when evaluating materials that have limited study data.

Computational Chemist Mihir Date, PhD, heads the clustering and read-across processes for evaluating fragrance safety at RIFM.

Q: What does clustering look like, and why do it?

Think of chemical clustering like arranging books in a library. If we shelve books of the same genre together and then shelve similar genres close to each other, it helps us locate relevant books more easily.

At RIFM, clustering a chemical inventory involves building groups of structurally similar chemicals (“clusters”) and then arranging together larger families made up of structurally related clusters.

In the absence of sufficient toxicity data on a compound of interest, clustering enables us to more efficiently identify structurally similar materials that may serve as read-across analogs (stand-ins for the compound of interest).

Q: How is clustering done?

We start with software programs like the OECD QSAR Toolbox (short for Organisation for Economic Co-operation and Development Quantitative Structure Activity Relationship Toolbox), which classifies chemicals according to their most reactive features, called “organic functional groups.” 

Each class is then subdivided into several clusters based on the other atoms attached to each chemical’s organic functional group, called “extended fragments.” RIFM performs this step using KNIME (Konstanz Information Miner), a data-mining and machine-learning tool.

After this automated process is complete, each cluster undergoes several reviews by chemistry experts to confirm the inclusion of chemicals in each cluster and to arrange the clusters according to their reactivity and toxicity

Q: How do scientists identify read-across analogs?

RIFM has developed a three-tiered system, which summarizes the structural similarity of possible read-across materials and provides a systematic way of searching for analogs that can be performed by scientists who may not be as familiar with chemical structures.

We have more fully described the tiered system in an upcoming peer-reviewed publication, which is currently “in press” and should be available soon.

Q: What makes clustering so useful?

Evaluating large groups of structurally different materials is challenging and time-consuming.

However, performing safety assessments on structurally similar materials helps scientists evaluate several chemicals at once. It also helps to identify read-across analogs that may be appropriate for the entire cluster, allowing for the evaluation of some target chemicals that may have little to no preexisting study data.

Clustering saves substantial time and effort—eliminating years that might be spent in the testing pipeline—and reduces the need for animal testing of chemicals that lack adequate studies.

Q: Can a read-across analog be used to predict a target chemical’s effect on multiple areas of human health?

Not necessarily. A read-across material that helps predict another material’s potential to affect the reproductive system may not be useful in predicting what impact that same material might have, if any, on human skin.

This is why there may be several different read-across materials identified in a RIFM safety assessment of a single target chemical.

Q: How reliable is chemical clustering?

Chemical clustering is a hypothesis and must be supported with data on several chemicals per cluster. For instance, alcohols and esters are data-rich and thus allow for highly accurate clustering and read-across. However, heteroaromatics are relatively light in data, and we may not be able to use this approach as much for some of them.

All read-across is approved by the Expert Panel for Fragrance Safety, an independent, international team of researchers and academics with no ties to the fragrance industry. Detailed read-across justifications are included in all RIFM safety assessments that employ read-across.

Q: Has clustering evolved with technological advancements?

Yes, it has come a long way. When chemists first started to develop clustering, they based structural comparisons on structural similarity scores, such as those provided by the Tanimoto index.

Structural similarity scores have since been augmented to account for many different features and behaviors of materials, like hydrogen bond donors vs. acceptorspolar vs. nonpolar surface areas, and 3D-shape alignment, which provides a more robust picture of the materials being compared.

Recently, researchers have started using artificial intelligence and machine-learning techniques to identify the critical properties of the chemical for clustering rather than a simple structural similarity score.

Read more about the Safety Assessment process and the innovative science that has helped RIFM save hundreds of thousands of animal lives.