Abstract:In life cycle assessment, characterization factors are used to convert the amount of the chemicals and other pollutants generated in a product's life cycle to the standard unit of an impact category, such as ecotoxicity. However, as a widely used impact assessment method, USEtox (version 2.11) only has ecotoxicity characterization factors for a small portion of chemicals due to the lack of laboratory experiment data. Here we develop machine learning models to estimate ecotoxicity hazardous concentrations 50% (HC50) in USEtox to calculate characterization factors for chemicals based on their physical-chemical properties in EPA's CompTox Chemical Dashborad and the classification of their mode of action. The model is validated by ten randomly selected test sets that are not used for training. The results show that the random forest model has the best predictive performance. The average root mean squared error of the estimated HC50 on the test sets is 0.761. The average coefficient of determination (R2) on the test set is 0.630, meaning 63% of the variability of HC50 in USEtox can be explained by the predicted HC50 from the random forest model. Our model outperforms a traditional quantitative structure-activity relationship (QSAR) model (ECOSAR) and linear regression models. We also provide estimates of missing ecotoxicity characterization factors for 552 chemicals in USEtox using the validated random forest model.

FRAGMENT CONSTANT MODEL FOR PREDICTION OF EC50 OFORGANIC COMPOUNDS TO DAPHNIA

A fragment constant QSAR model for evaluating the EC50 values of organic chemicals to Daphnia magna.

A fragment constant QSAR model for evaluating the EC50 values of organic chemicals to Daphnia Magna☆☆Funding was provided by The National Scientific Foundation of China [49971070, 40024101, 40031010].

PREDICTION OF BIOCONCENTRATION FACTOR OF NON-POLAR ORGANIC COMPOUNDS BY FRAGMENT CONSTANT REGRESSION

FRAGMENT CONSTANT MODEL FOR Koc ESTIMATION

Fragment Constant Method For Prediction Of Fish Bioconcentration Factors Of Non-Polar Chemicals

Quantitative structure activity relationship model for predicting the depletion percentage of skin allergic chemical substances of glutathione.

A Comparison of the Fragment Constant and Molecular Connectivity Indices Models for Normalized Sorption Coefficient Estimation.

Estimation of Organic Carbon Normalized Sorption Coefficient (K-Oc) for Soils Using the Fragment Constant Method

A fragment constant method for estimation of bioconcentration factor of hydrophobic chemicals

Predicting of BCFs in Fish for Organic Chemicals Using Fragment Constant Metho D

[Estimation of LC50 of Chemicals to Rainbow Trout by Fragment Constant Method].

QSAR MODELING OF BIOCONCENTRATION FACTORS IN FISH BASED ON FRAGMENT CONSTANTS AND STRUCTURAL CORRECTION FACTORS

Robustness of a koc prediction model based on fragment constants

A Modified Diffusion-Limited Cluster Aggregation Model for Accurate Prediction of the Coagulation and Fragmentation Process in Nanoparticle Suspension

Comparison of KOC Prediction Models on the Based of Fragment Constants and Molecular Connectivity Indices

Prediction of the Bioconcentration Factor of Pcbs in Fish. Using the Molecular Connectivity Index and Fragment Constant Models

A QSAR Model for Predicting Toxicity (LC50) to Rainbow Trout.

Estimate Ecotoxicity Characterization Factors for Chemicals in Life Cycle Assessment Using Machine Learning Models.

Reference dose prediction by using CDK molecular descriptors: A non-experimental method

Predicting Both Lower and Upper Flammability Limits for Fuel Mixtures from Molecular Structures with Same Descriptors