Adapting to a Shifting Regulatory Landscape by Harnessing Science-Based AI for Sustainable Ingredient Substitution
The European Union (EU) is at the forefront of a significant shift in chemical regulation, moving from targeting individual compounds to considering groups of chemically similar substances12. This approach aims to address the broader environmental and health impacts of chemical pollution more effectively by circumventing the oft-practiced 'regrettable substitution' method of selecting a less-scrutinized structural derivative that ultimately ends up having a similar toxicity and environmental impact as the replaced component. This new approach to chemical regulation targets hazardous chemicals like bisphenols, used to enhance material strength; phthalates, which add flexibility to materials; alkylphenol ethoxylates (APEO) used as surfactants in coatings for improved wettability; and per- and polyfluoroalkyl substances (PFASs), known for their heat and chemical resistance in packaging and fire-retardancy in fire-fighting foams, textiles, and other consumer goods. While the United States Environmental Protection Agency (EPA) continues to regulate individual chemicals at the federal level, some states are taking a more aggressive approach (akin to the proposed EU regulations) to prevent ‘regrettable’ substitutions. The multi-nation regulatory shift represented by the EU's proposed action underscores a global trend towards more comprehensive chemical management that puts public and environmental safety at the forefront. For consumer and industrial product companies, the shift towards potential group bans of similar chemistries will necessitate innovative strategies to navigate the complex and evolving regulatory landscape, ensuring compliance to safeguard public health and the environment.
Understanding The Role of Molecular Similarity
At the heart of these regulatory changes is the concept of molecular similarity. Traditionally, when a specific chemical was banned, companies would seek out chemically similar alternatives that could replicate the original's function. This practice, while effective under previous regulations, now carries inherent risks. The EU's grouping of similar chemicals means that substituting one banned substance for a similar, yet still legal, alternative might no longer be a viable strategy in the long-term. The product development cycle, from research to commercialization, often spans years. Investing in the formulations using chemically similar substitutes can lead to significant operational inefficiencies and financial losses if these substitutes are later banned. This necessitates a strategic pivot in product development, urging companies to reconsider their reliance on molecularly similar chemicals and explore more distinct alternatives to mitigate the risk of future regulatory restrictions and associated costs. Exploring this frontier of structurally dissimilar but functionally similar space is a significantly more challenging prospect, since it deviates from a fundamental concept of chemical intuition: structure-function correlation and the concept of like ~ like.
Leveraging Science-Based AI to Overcome Regulatory Hurdles
At the heart of tackling the challenges posed by new regulatory frameworks is Science-Based Artificial Intelligence (SBAI), a novel technology from NobleAI. SBAI combines scientific knowledge and principles with optimally selected ML algorithms. One aspect of SBAI technology is the advanced techniques of molecular featurization. This critical process transforms complex molecular structures into a vectorized numerical format, making them accessible for machine learning models. Techniques such as molecular fingerprints and graph-based representations serve as the linchpins in this process, allowing computational algorithms to analyze molecules at an atomistic level3.
Machine learning models can utilize these detailed molecular representations, by encoding specific chemical features, bonds, and spatial arrangements. One simplified example of a molecular featurization approach is shown in Figure 1 where fragments of a toxic PFAS material are encoded into a vector representation. Molecular featurization forms an important aspect for SBAI models, which can be adept at identifying structurally dissimilar, less toxic alternatives to banned chemical groups. Encoding molecules and projecting them into two-dimensional spaces can facilitate the identification of toxic chemical clusters with SBAI as illustrated in Figure 2. Alternatives are then selected based on their proximity within this space, with those further from toxic clusters being deemed less likely to face future bans.
The Complexity and Challenges of Substituting Chemicals
The principle "like dissolves like" resonates deeply within the realm of chemistry, guiding the exploration and innovation in molecular design. Historically, this principle has steered efforts towards finding molecules with similar structures, as these often share behaviors and functionalities, with researchers looking to see how incremental changes in structure impart subtle changes to properties to tune and control behavior. However, the challenge of product development may now be shifting toward identifying structurally dissimilar chemistries that nonetheless fulfill similar roles, without compromising on performance within their specific application areas coupled with increased scrutiny on understanding toxicity and environmental effects.
SBAI models and their advanced molecular featurization techniques come into play as powerful tools to navigate this complex optimization task as shown in Figure 3. The criteria for successful substitution have become multifaceted, focusing on:
- Ensuring Benign Toxicity: Selecting alternatives that are safe for both humans and the environment.
- Maximizing Structural Dissimilarity: Choosing chemical structures or functional groups that are different from banned groups to minimize the risk of future regulatory restrictions.
- Maintaining Functional Integrity: Finding substitutes that offer comparable or enhanced performance for the intended application while maintaining appropriate cost controls.
Leveraging SBAI models equipped with molecular featurization allows for the utilization of optimization algorithms that can pinpoint the ideal balance among these critical requirements. This not only accelerates the pace of product development by identifying viable replacements swiftly but also significantly reduces the long-term risk of adopting formulations that could later be banned due to structural similarities with currently prohibited substances. NobleAI’s SBAI models can be deployed on the NobleReactor platform to drive the discovery of the most sustainable ingredient substitution that has the lowest risk of being banned in the future. Objectives can be defined around structural dissimilarity to banned chemicals, toxicity minimization, and property targets to maintain product performance. Scientists can use NobleReactor to automate the exploration of all possible chemical formulations or combinations, identifying compounds that could replace toxic ingredients without sacrificing performance or quality.
In this new regulatory era, SBAI models will emerge as a pivotal strategy for ensuring regulatory compliance and product reformulation, empowering companies to stay ahead of the curve. With the help of SBAI, the path towards innovation, safety, and compliance will become more clear, promising substantial benefits for public health, environmental stewardship, and the resilience of businesses navigating these changes.