MS2BCF: Prediction of bioaccumulation factor from high-resolution mass spectrometry data using machine learning

A chemicals persistence in the environment, its ability to bioaccumulate in organisms, and its toxicity are the core principals to the definition of persistent organic pollutants (POPs) as outlined by the United Nations Stockholm Convention. However, the ability to characterise a chemical as a POP is hindered by the vast number of chemicals detected in the global environment. Acquisition of high-resolution mass spectrometry (HRMS) data from environmental and biological matrices can result in tens of thousands of multidimensional regions of interest (features). Suspect and non-target screening strategies aim to identify and characterise these features based on known and predicted parameters, such as the mass to charge (m/z) and retention time. However, it remains very time consuming and requires significant compute resources to process and evaluate the results from each feature. Therefore, prioritization strategies are being developed to help highlight features of importance for further investigation. Recent advances in machine learning assisted workflows are able to predict a features toxicity, based solely on its measured mass spectrum, significantly improving the prioritization workflow. This research project aims to develop the first analytical tool for the estimation of the bioaccumulation factor (BCF) from HRMS data and propose a methodology to assess the risk of unidentified chemicals in the environment. The model will combine empirical evidence of BCF and mass spectrum collected from a variety of databases and other peer-reviewed sources. The model will then be validated with a retrospective analysis of unknown contamination in the environment to provide an evaluation of the predicted BCF for the first time. The outcomes of this project will positively contribute to the assessment of impact of novel entities in the environment, an important global boundary.