Uncovering sustainable personal care ingredient combinations using scientific modelling

This paper proposes a pioneering approach using predictive modeling and digital simulation services to identify high-performing, sustainable natural ingredient combinations that can effectively replace synthetic components in personal care formulations amid tightening regulations and consumer demand.

The Gist

Imagine you are a master chef trying to recreate a famous, delicious dish. But there's a catch: the original recipe calls for a secret ingredient that is now banned by the health department (like a specific type of artificial oil or silicone) because it's bad for the environment.

Your goal? You must create a new dish that tastes, feels, and looks exactly like the original, but using only fresh, natural, and sustainable ingredients. And you have to do it quickly, without spending a fortune on trial and error.

This is exactly the challenge facing the personal care industry (shampoos, lotions, makeup), and this paper describes how a team at BASF used computer magic to solve it.

Here is the breakdown of their solution in simple terms:

1. The Problem: The "Needle in a Haystack"

For years, companies used synthetic ingredients like silicones (which make hair feel silky) and mineral oils (which make skin feel smooth). They worked great, but they are now being banned or shunned because they don't break down in nature.

Traditionally, a "formulator" (the person who mixes the products) would have to guess and check. They would mix Ingredient A with Ingredient B, test it, hate it, mix Ingredient C with D, test it, and so on. This is like trying to find a specific needle in a massive haystack by feeling every single piece of hay with your eyes closed. It takes forever, costs a lot of money, and you might still miss the perfect needle.

2. The Solution: The "Digital Taste-Tester"

Instead of guessing, the authors built a digital simulator. Think of this as a super-smart computer chef that has read every recipe book ever written and can predict exactly how ingredients will taste and feel before you even mix them.

They used two main tools:

• The Math Model (The "Recipe Calculator"): This looks at the basic properties of natural ingredients (like how oily they are or how thick they are) and uses math to figure out how to mix them to mimic the banned synthetic ones. It's like knowing that if you mix 1 cup of oil A with 2 cups of oil B, you get the exact same texture as the banned silicone.
• The AI (The "Experience Learner"): For more complex mixtures, they used Artificial Intelligence. Imagine an AI that has tasted millions of different lotions and shampoos. It learns patterns that humans might miss, like "When you add this specific plant extract, it stops the foam from getting too bubbly."

3. The Results: Finding the Needle

The paper shows three specific examples where this digital chef succeeded:

• Replacing the "Silicone Slip": They needed to replace a banned silicone (D5) used in makeup removers. The computer suggested a specific mix of three natural oils (from coconut and plants). When they tested it, it felt and cleaned just as well as the banned silicone.
• Replacing the "Silicone Smoothness": They needed to replace a lighter silicone used in face serums. The computer found a natural blend of plant-based carbonates and ethers that felt identical to the silicone, without making the skin feel sticky or foamy.
• Replacing the "Synthetic Suds": They needed to replace a synthetic foaming agent in shampoo. The computer suggested a mix of natural sugar-based and amino-acid-based surfactants. The result? A shampoo that foams just as beautifully as the synthetic version but is 100% natural.

4. Why This Matters

The authors compare their method to how our brains work. Our brains use "shortcuts" to make quick decisions so we don't waste energy thinking about every single detail. This computer system does the same thing for scientists.

The Big Takeaway:
Instead of spending years and millions of dollars testing thousands of random combinations in a lab, formulators can now use this digital platform to get a shortlist of the best, most sustainable recipes instantly. They only have to test the top few suggestions in the real world.

In a nutshell: This paper is about using AI and math to turn the impossible task of "finding a natural replacement for a banned chemical" into a quick, easy, and successful recipe for a greener future.

Technical Summary

1. Problem Statement

The personal care industry faces a critical convergence of regulatory pressure, consumer demand, and economic constraints:

• Regulatory Bans: The EU is banning specific cyclic siloxanes (D4, D5, D6) and restricting mineral oils due to their persistence, bioaccumulation, and toxicity (PBT/vPvB properties).
• Performance vs. Sustainability: Consumers demand "natural" formulations that match the performance (sensory feel, stability, efficacy) of established synthetic ingredients (silicones, mineral oils, synthetic surfactants) without a significant cost increase.
• Formulation Complexity: Replacing synthetic ingredients is a "needle in a haystack" problem. Traditional formulators rely on trial-and-error, which is time-consuming, costly, and risks missing optimal natural combinations.
• The Challenge: There is an urgent need for a systematic, rapid method to identify natural-based ingredient combinations that can serve as "drop-in" replacements for banned or restricted synthetics without compromising product quality.

2. Methodology

The authors propose a scientific modelling and simulation-based digital service platform that utilizes mathematical modelling and Artificial Intelligence (AI) to predict ingredient performance. The methodology is structured in three tiers:

A. Data Decomposition (Signal vs. Noise)

The foundation of their approach (Eq. 1) treats property measurements as a linear summation of a signal (intrinsic, fixed properties) and noise (experimental error, environmental factors).

• Goal: Develop models that capture the signal while minimizing overfitting to noise.
• Principle: Models are validated on unseen data to ensure they predict intrinsic properties rather than random experimental variations.

B. Simple Mathematical Modelling (Ingredient Level)

For individual ingredients or simple mixtures (e.g., emollients), the authors use mixing models to predict the properties of a mixture ($prop_{mix}$) based on the properties of individual components ($prop_i$) and their ratios ($x_n$).

• Equation: $prop_{mix} = f(x_n, prop_n)$
• Function: The function $f$ can be polynomial, exponential, or logarithmic.
• Application: Used to predict physico-chemical properties (viscosity, surface tension, polarity) and sensory attributes (powdery, waxy, silicone feel).
• Accuracy: These models achieved >95% accuracy on untrained test data.

C. Target Optimization

The system solves an optimization problem (Eq. 3) to find the specific ingredient ratios that minimize the error ($\Delta$) between a natural mixture and a synthetic benchmark (e.g., D5 silicone).

• Objective: Minimize $|prop_{benchmark} - prop_{mix}|^2$.
• Output: The system iterates through all possible combinations to identify the "needle in the haystack"—the specific natural blend that mathematically matches the synthetic benchmark.

D. Machine Learning (Formulation Level)

For complex formulations involving multiple ingredient types where simple mixing models fail, the authors employ Machine Learning (ML) and Graph Neural Networks (GNN).

• Equation: $prop_{formulation} = f_{ML}(x_n, prop_{mix})$
• Inputs: Ratios and properties of all ingredients in the final formulation.
• Outputs: Predictions for stability, rheology, and complex sensory profiles.
• Tools: Built using Python (Scikit-learn) and advanced GNNs to handle complex chemical interactions.

3. Key Contributions

The paper introduces two specific digital simulators developed by BASF:

1. Emollient Simulator: Designed to replace synthetic oils (silicones, mineral oils) with natural blends.
2. Surfactant Simulator: Designed to replace synthetic surfactants (sulfated, EO-based) with non-sulfated, non-EO natural alternatives.

The core contribution is the shift from heuristic trial-and-error to computational intelligence, allowing formulators to screen thousands of theoretical combinations instantly and test only the most promising candidates.

4. Results and Case Studies

The authors validated their models through three specific case studies:

Case 1: Replacement of Cyclopentasiloxane (D5)

• Benchmark: D5 (used in makeup removal).
• Solution: A ternary natural mixture of coco-caprylate/caprate, undecane/tridecane, and dicaprylyl ether (ratio 1:2:1).
• Outcome: The predicted mixture matched D5 in physico-chemical properties and monadic sensory attributes. In a simplex makeup removal formulation, the natural blend performed identically to the D5 baseline in cleansing efficacy and sensory feel.

Case 2: Replacement of Dimethicone (5 cSt)

• Benchmark: Linear Dimethicone (5 cSt) used in skin serums for sensory feel and anti-foaming.
• Solution: A ternary natural mixture of dicaprylyl carbonate, dicaprylyl ether, and undecane/tridecane (ratio 1.7:3.2:1).
• Outcome: The natural blend matched the silicone's viscosity and sensory profile. Crucially, the formulation maintained anti-foaming properties, a common failure point in silicone-free formulations where natural thickeners often cause unwanted foaming.

Case 3: Replacement of Synthetic Surfactants ($\alpha$-olefin sulfonate)

• Benchmark: A mixture of synthetic $\alpha$-olefin sulfonate and cocamidopropyl betaine (3:1 ratio).
• Solution: A natural, non-sulfated, non-EO blend of lauryl glucoside and sodium cocoyl glutamate (1:1 ratio).
• Outcome: The natural surfactant blend demonstrated comparable foaming performance, bubble formation, and foam elasticity to the synthetic benchmark.

5. Significance

• Accelerated Innovation: The digital approach drastically reduces the time and cost associated with reformulation, enabling brands to meet regulatory deadlines (e.g., EU 2027 ban) rapidly.
• Genuine Sustainability: It moves beyond "greenwashing" by providing scientifically validated, high-performance natural alternatives that do not require compromising on consumer experience.
• Scalability: The methodology is not limited to the examples shown; it can be applied to a wide range of ingredients including isododecane, mineral oils, squalene, and various surfactants (SLS, taurates, etc.).
• Industry Transformation: By empowering formulators with AI-driven "short-cuts" (similar to human heuristics), the paper advocates for a fundamental shift in the personal care industry toward data-driven, sustainable development.