Unveiling Chemical Fingerprints: Applications of Advanced Machine Learning for Environmental Forensics in Plastic Pollution

Keywords: microplastics, non-targeted fingerprinting, clustering analyses, supervised machine learning.

What are environmental forensics and chemical fingerprints?

Looking at the term “fingerprints” or “forensics” in the title of this blog, it probably reminds you about a detective TV show that you have seen or a book that you have read, for example, Brooklyn 99 or about the famous detective Sherlock Holmes. In these scenarios, the detective will search for a fingerprint of the criminal and use it to track down the identity of the one who committed the crime. Similarly, in environmental forensics, the goal is to link local pollutions to responsible local polluters before more harms are done.

Application of environmental forensics in plastic pollution?

In my research, I focus on application of environmental forensics in plastic pollution. Plastic pollution has been repeatedly announced as a massive global issue and ecotoxicological research has identified thousands of toxic substances leaching from plastics in the environmental. Yet, since there are thousands of different chemicals that are being used in the production step of various types of plastic products that we use these days, we are lacking a tool to effectively analyze such big data to do source-tracking and developing mitigation efforts.

Why environmental source tracking of plastic contaminants is so difficult?

A prominent challenge of chemical forensic studies, which seek to identify the source of a chemical/mixture of interest, is that 10−100 s of thousands of chemicals exist in the environment. However, current provincial and federal monitoring programs screen for mere hundreds of compounds (which you can find here (external link)  and here (external link) ) are ignoring ∼>99.9% of all chemical features present (i.e., the chemical space), so the chances of identifying the specific chemicals driving adverse environmental phenomena are small. Therefore, robust chemical forensic strategies are needed for routine monitoring that utilize broader chemical space. Moreover, when it comes to plastic pollution, we have challenges where plastic and its associated contaminants can 1) be transported far from original source of pollution, 2) contains a diverse suite of chemical, both comes from the polymer backbone and added plastic additives during the manufacturing of plastic products and 3) these plastic contaminants might be exposed to environmental weathering and hence alter in their chemical fingerprints. With these challenges, just using target or suspect screening approach, although we can confirm which toxic chemical that come from these sources but not exactly able to do source pointing. All these challenges further emphasize the need for a better data processing approach.

In my project, I developed a computational fingerprinting workflow for environmental forensics of plastic with several goals in mind:

1. What are the chemical fingerprints and their patterns in locally used plastic products?
2. How does concentration of chemical fingerprints in complex contaminant mixtures correlate with the sources of contaminant?
3. What is the prediction accuracy of the automatic identification of the sources of contaminants?

List of representative chemical fingerprints and their functions, associated industry and plastic products

With great concerns, we found that the plastic products that have the highest concentration of plastic additives are food contact materials (Food packaging, bottle caps) and children’s toys (plastic toy balls). Moreover, there is also a lack of information on the functions and associated industry of these representative chemical fingerprints

Here we see that a perfect prediction accuracy where plastic product type in the test data (environmental plastics) was successfully tracked back to their store-bought equivalents with high degree of certainty for Food packaging, plastic cups and fishing bait trays (with value of 1 as 100% certainty). With this result in mind, we can confirm that the chemical fingerprints that was identified are stable enough and can be considered as representative for environmental source tracking of these plastic products

The Next Steps

Moving forward, the developed computational fingerprinting approach can provide a robust workflow for analytical chemist to come up with their own fingerprints for their specific contaminants of interest. Product or producer-specific information on microplastic and plastic additive emissions generated from this workflow can help policymakers plan chemical management strategies to set regulatory parameters for microplastics; as well as inform consumer choices leading to a potential reduction in emissions and exposure to hazardous contaminants. Being able to identify the producers or specific product types that contribute to local environmental microplastic loads can inform regulatory action for the targeted reduction of microplastic emissions and potential cost recovery from polluters. Moreover, localized information on emission sources for specific microplastics and associated additives can inform the adaptation of wastewater treatment techniques to increase the removal efficiencies for these specific contaminants.