In ZeroPM pieces #26 Sivani Baskaran, a Postdoctoral Researcher at Norwegian Geotechnical Institute (NGI), presents how data science approaches can be used to assess mobility, as part of #PMT Assessment.
You can download the slides of this presentation here: zenodo.org/records/13946235
Experimental and estimated KOC data was extracted from eChemPortal using the property search. Experimental partition ratios (KOC, KOW, and DOW) and pKa data submitted to REACH dossiers was obtained from the OECD QSAR Toolbox. Additional experimental data was also obtained from the training and validation data used by different property prediction models. The chemical identifiers associated with each of the entries was then processed using the Cleanventory process to obtain the associated InChI value. The KOC is estimated using several existing tools, such as the OPERA model which uses a k-nearest neighbour approach, the US EPA’s EPI Suite KOCWIN models which includes two fragment-based models using the KOW and molecular connectivity index, multiple poly-parameter linear free energy relationships (PP-LFERs), and single-parameter linear free energy relationships (SP-LFERs). These models are parameterized for neutral organic compounds and so may not work so well for ionizable chemicals. Any chemical with a pKa is thus considered ionizable. pKa can be predicted using OPERA and ACD/Labs. KOC values are predicted for all 18M+ SMILES using all available predictions.
As a result of the data extraction and prediction process, we have millions of KOC values associated with the chemicals on the global chemical inventory. The initial analysis of the experimental KOC data indicates that the data reported and extracted from REACH dossiers using the eChemPortal is easily misinterpreted because the nuance and details included in the original dossiers is not included in the extracted files. As a result, it is difficult to identify errors without referring directly to the original REACH dossier. In some instances, the extracted data is indicated to be an experimental value; however, within the original dossier the value is noted to be a prediction. While these extreme cases are easily identified, we can not rely on the data curation of the eChemPortal to filter experimental and estimated data. In addition, the extracted data set includes repeated data which is not filtered out during the data extraction process by eChemPortal. All duplicate values will bias the analysis using the available data. Therefore, the data is cross-referenced with data from the OECD QSAR Toolbox. If property values for a given chemical matched up to 2 decimal places, they were considered duplicated and removed from the dataset. Data obtained directly from peer-reviewed literature sources are considered to be the most reliable, followed by experimental data from the QSAR Toolbox. Finally, data from the REACH Dossiers and estimated data from This process resulted in a dataset of highly reliable property data from published literature and REACH dossiers and less reliable data from eChemPortal.
Much of the experimental data and data obtained from the OECD's eChemPortal and QSAR Toolbox are not reported as point values. These values indicate the value of the property will be above or below a particular threshold or between a particular set of numbers. We use a Bayesian analysis approach to determine a consensus experimental value (and its uncertainty) based on all available data. By including all data in our approach as is, we can use a maximum of data for analysis and can obtain a better understanding of the uncertainty of the final value, and for KOC values in general.
In previous work, we showed in a proof of concept that this Bayesian analysis approach to deriving consensus values can work to determine a probability distribution for the KOC of individual chemicals. With the lack of reliability in "experimental" values extracted from the eChemPortal and a distinct lack of meta data we are currently applying this approach to predict and assess the probability of each of the 340,000 chemicals in the global chemical inventory as being mobile, very mobile, or not mobile. Consensus based physical-chemical property data is preferable to using a single estimated or experimental value where reliable and applicable models are available.
In the ZeroPM pieces video series, individual researchers in the project will tell you about one piece of the ZeroPM puzzle they are working on. These were filmed as mini-lunch time seminars within the ZeroPM project, and are being released in different seasons via this youtube channel.
For more info on ZeroPM, please visit our website at www.zeropm.eu.
#PFAS #PMT #zeropollution #environmentalchemistry #hazard #risk #bioassays #endocrinedisruptors
ZeroPM has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 101036756