My interest is finding ways to study complex chemical systems and develop tools to make it easier for others. During my studies I developed methods to identify textile fibres and natural dyes to characterize different historical and archaeological findings. In my first postdoc, MS2Tox was developed for prioritizing more toxic compounds in environmental samples.
Now in VEO group, my aim is to find ways to analyse and compare metabolomes, bypassing identification of individual compounds, in order to use the full potential of LC-HRMS obtained data.
During my first postdoc in Stockholm University a machine learning based prediction model named MS2Tox was developed to estimate LC50 values for chemicals that are not fully identified from environmental samples. Due to abundance of LC-HRMS features, level 1 identification of all the compounds in sample is often not possible. Level 1 identification needs confirmation with standards and testing thousand of possible matches can be extremely time and resource consuming. In order to concentrate on compounds that possibly contribute more to toxic effect of the sample, prioritization of compounds is much needed. Thus MS2Tox, which relies on chemical characteristics, like mass and structural fingerprints, can be used to prioritize more important features in LC-HRMS dataset. Besides chemicals toxicity, also its exposure (concentration) is important. For concentration estimations in non-targeted analysis, in the same Kruvelab group, MS2Quant has been developed by Helen Sepman. Combining toxicity and concentration information of all the detected compounds from sample can give more realistic overview of the environmental condition, than common methods of monitoring specific target compounds. More about MS2Tox and MS2Quant from https://doi.org/10.1021/acs.est.2c02536External link and https://doi.org/10.1021/acs.analchem.3c01744External link .
