Possible starting date: From mid September 2022
With the development of high-resolution mass spectrometry and nontargeted screening workflows, we are able to detect and process tends of thousands of unknown and potentially harmful environmental pollutants in aquatic samples. However, since even the most sophisticated data analytics workflows require manual verification for tentative identification and quantification, we have to prioritize which compounds are most relevant in different studies. Such prioritization strategies are often based on the measured areas or frequencies of occurrence, both of which lack toxicological relevance so critical in the context of environmental pollution. In order to append toxicological relevance to HRMS analysis, machine learning methods (ML) are being developed to predict the toxicity of unknown features based on their MSMS spectra and thus prioritize the compounds based on their potential toxic effect.
In this study, we want to focus on the analytical validation of MLinvitroTox, an ML tool under development at Eawag for the prediction of toxicity fingerprints for unknown HRMS features based on their MSMS spectra. We aim to deploy MLinvitroTox on wastewater samples for prioritization of potentially toxic species towards tentative identification and quantification, followed by confirmation/rejection with authentic standards. In this work, we will collect and analyze wastewater samples as well as use the “digitally frozen” sample repository already present at Eawag. This work will help to establish the validity of the proposed approach as part of the EXPECTmine project (Mining toxicity and HRMS data for linking exposures to Effects: https://www.eawag.ch/en/department/uchem/projects/translate-to-english-expect).
The student will gain skills in working with analytical techniques in general and in particular ultra-high-performance liquid chromatography-tandem high-resolution mass spectrometry (UHPLC-HRMS/MS). More specifically, the student supported by the supervisor will collect and analyze wastewater samples, perform targeted, suspect, as well as nontargeted analyses, deploy MLinvitroTox on the results, identify potentially toxic features of interest, and attempt to confirm their identity analytically. The work will combine fieldwork, analytics, machine learning, and cheminformatics.
Supervised by Dr. Kasia Arturi, Prof. Dr. Juliane Hollender.
Contact: firstname.lastname@example.org, email@example.com