Für Mensch & Umwelt LIFE COMBASE workshop on Computational Tools for the Assessment and Substitution of Biocidal Active Substances of Ecotoxicological Concern Screening and prioritisation of substances of concern: A regulators perspective within the JANUS project Jürgen Arning Section IV 2.3 Chemicals German Environment Agency (UBA) 19. October 2017, Milan, Italy
JANUS Joining environmental, ecotoxicological and toxicological Assessment of chemical substances with Non-testing methods within a Unified Screening 2
JANUS Project Team Thank you! 3
Background Role of MS under REACH: Identifying substances for regulation Prioritisation of substances of very high concern (SVHC) Potential SVHC candidates: CMR cat. 1 or 2 PBT/vPvB other substances of equivalent concern, e.g. Endocrine Disruptors Role of UBA: screening/identifying SVHC with respect to the environment 4
Identification of SVHC substances Problem: Poor data quality (e.g. incomplete documentation of QSARs and waiving) Difficult substances: e.g unkown and/or complex mixtures, highly hydrophobic chemicals, non-standard bioaccumulation e.g. via protein binding Less information in upcoming registration period (1 100 t) Identification of SVHC candidates based on in silico methods! 5
Aims of the JANUS project Main Objectives To set up a conceptual scheme suitable for identification and prioritisation of SVHC To implement an in silico tool for SVHC assessment Related Objectives To integrate and broaden a battery of in silico models for hazard assessment To verify the performance on a set of compounds 6
Where we are The results of PROMETHEUS Model successfull in separating PBT vs non-pbt Integration of many in silico models and rules Consensus models Implemented within VEGA Possible improvement with ED assessment and human health toxicology 7
Where to go The JANUS concept P Ready biodegradability Quantitative persistence - Water - Sediment - Soil Microbial biodegradation P B T Score (1 = PBT; 0 = npbt) + ED B BCF Log Kow BCF integration Metabolism in fish CMR Cancerogenesys Mutagenicity Reprotox Metabolism in mammals T Chronic Fish toxicity Chronic D. magna toxicity Chronic algae toxicity Metabolism in fish CMR ED Endocrine disruptors 8
The JANUS work packages 9
Example workflow: Fish toxicity Main parameters: Experimental values (single/multiple/ agreement) Predicted values (single/multiple/ agreement) ACR (high/low/presence) 10
The new workflow for T assessment Aquatic toxicity ECOTOX Aquire Aquatic Japan MoE Continuous data Acute Chronic PROMETHE US (ECHA+TB) # chemicals for algae 315 408 # chemicals for daphnia # chemicals for fish 331 428 306 35 94 More data from ECHA 11
The new workflow for T assessment Aquatic toxicity CURATED DATASETS ACCORDING TO EXP CONDITIONS, CHEMICAL STRUCTURE, EXP VALUES, CHEMICALS WITH TOX VALUE > WATER SOLUBILITY, Continuous data Acute Chroni c # chemicals for algae 315 408 # chemicals for daphnia 428 306 # chemicals for fish 331 94 PROCESSING BEFORE MODELLING BOX-COX TRANSFORMATION (ʎ optimization) EXCLUDED CHEMICALS OUTSIDE THE RANGE MEDIA±3*ST SMILES NORMALIZATION MODELLING Neutralization (D. magna and fish) Ionization (algae) Excluded chemicals whose SMILES changes from ph 7.5 to 8.1 SPLIT TRAIN 80% - TEST 20% (sampling: maxdissim according to exp values, PC1 and PC2) 12
The new workflow for T assessment Aquatic toxicity NORMALIZED SMILES Dragon 7 descriptors calculation Search for the most correlated descriptors to the endpoint and elimination of chemicals that give missing values for them Feature selection (packages: VSURF, gaselect) Random Forest Hyperparameter tuned with bootstrap 13
The new workflow for T assessment T - SCREENING Aquatic toxicity ENDPOINT TRAINING SET VALIDATION SET Without AD With AD R2 MAE RMSE R2 MAE RMSE R2 MAE RMSE coverage Algae EC50 0.95 0.46 0.59 0.56 1.02 1.33 0.65 0.94 1.18 0.79 NOEC 0.95 0.56 0.74 0.58 1.36 1.73 0.63 1.29 1.66 0.93 Daphnia magna EC50 0.94 0.49 0.64 0.56 0.99 1.31 0.69 0.84 1.09 0.84 NOEC 0.95 0.42 0.56 0.74 0.83 1.13 0.78 0.80 1.07 0.81 Fish LC50 0.95 0.27 0.37 0.65 0.68 0.87 0.65 0.64 0.83 0.81 NOEC 0.96 0.54 0.68 0.74 1.75 2.45 0.76 1.79 2.54 0.89 T - ASSESSMENT 14
The new workflow for T assessment Aquatic toxicity More fish chronic exp data from ECHA? Integration of the models to obtain the final assessment for T (ecotoxicity) 15
The new workflow for T assessment Mutagenicity consensus model CMR Qualitative prediction on Salmonella typhimurium (Ames test), applying a consensus-based model approach on the 4 QSARs in VEGA The dataset was enlarged with data from the Japan Health Ministry We further tested the consensus and tuned the strategy to improve the outcome, leading to the current version (1.0.2) in VEGA Models under development. Zebrafish embryo Slope Factor models for carcinogenicity Slope Factor oral model Slope Factor inhalation model In regression In classification More than 700 molecules with Slope Factor values from: IRIS database Risk Assessment Information System ISS-INAIL TERA 2.5 Data pruning (no duplicates, inorganics and mixtures and neutralization of salts) 16
The new workflow for T assessment Mutagenicity consensus model CMR 17
The new workflow for T assessment Mutagenicity consensus model CMR 18
The new workflow for T assessment Mutagenicity consensus model CMR 19
Integration of ED assessment Estrogen Receptor - ER Source of data Models Description Androgen Receptor - AR OECD QSAR Toolbox profilers (Estrogen Receptor Binding, DART scheme v. 1.0, rter Expert System v.1 USEPA) Estrogen Receptor-mediated effect (IRFMN/CERAPP) Estrogen Receptor Relative binding model (IRFMN) SAR Classification and Regression Tree (CART) Structural Alerts extraction Source of data Models Description Androgen Receptor binding potential (IRFMN/CoMPARA) Androgen Receptor binding potential (SARpy-IRFMN/CoMPARA) Decision Tre Structural Alerts extraction 20
Integration of ED assessment DART Source of data Models Description Developmental Toxicity model (CAESAR) Developmental/Reproductive Toxicity library (PG) SARpy developmental toxicity Random Forest Rule-based Structural Alerts extraction Tentative DART & ED workflow AR or ER pos AR or ER pos AR or ER neg DART pos DART neg DART pos AR or ER neg DART neg Lower concern Higher concern (apical effects pos + ED related assays pos) Medium concern (no apical effect on DART underlying uncertainty: other non DART effects not covered) Medium concern (no apical effect on DART underlying uncertainty: other mechanisms not covered) 21
Outcome for prioritization and assessment Ranking DES (PBT) = 0.75 Ranking DES (PBT) = 0.19 DES: Global Desirability Index 22
Thank you for your attention! Jürgen Arning German Environment Agency - Umweltbundesamt (UBA) Section IV 2.3 Chemicals Wörlitzer Platz 1, 06844 Dessau +49-340 - 2103-2171 Email: juergen.arning@uba.de www.uba.de 23