Registration Dossier

Environmental fate & pathways

Adsorption / desorption

Currently viewing:

Administrative data

Link to relevant study record(s)

Reference
Endpoint:
adsorption / desorption
Remarks:
other: KOCWIN v. 2.0 prediction.
Type of information:
(Q)SAR
Adequacy of study:
key study
Reliability:
2 (reliable with restrictions)
Rationale for reliability incl. deficiencies:
other: In domain and well documented prediction using analogues for validation.
Justification for type of information:
QSAR prediction: migrated from IUCLID 5.6
Principles of method if other than guideline:
KOCWIN v. 2.0 prediction.
GLP compliance:
no
Remarks:
Not applicable to QSAR.
Type of method:
other: KOCWIN v. 2.0 prediction.
Media:
other: N.A. Majority of training set data is on water-soil partitioning.
Type:
log Koc
Value:
2.65
Temp.:
20 °C
Remarks on result:
other: QSAR value KOCWIN Kow method.

Result from the Domain assessment:

KOCWIN does not provide explicit information about applicability domain of the model. The applicability domain can be derived from the training data set and the fragments with correction. If a given unknown consist of fragments that are not aliphatic or aromatic and do not have a correction factor listed they have to be considered out of domain.       

i.     Parametric Domain: Not specified by developer.

Log Kow: minimum–maximum of in the training data:   
               -2.11 < Log Kow< –9.1           
Experimental Kowof the unknown (column method): 3.24         

MW: minimum–maximum of molecular mass in the training data:         

               32 < MW < 665          [g/mol]
MW of unknown: 326.4 [g/mol]

=>In domain

ii.     Structural Domain: Not specified by developer. Assessed as all molecules with either aliphatic or aromatic structural fragment or with a fragment with a correction factor defined for the model.  
Fragments of the unknown:

-         Aliphatic

-         Aromatic

-         Epoxy (Ether). A series of Epoxy are represented in the training set (see analogues)

=>In domain

iii.     Mechanism domain: The model does not define a mechanism of partitioning. However, the dominant driving force for partitioning of the molecules in the training set is hydrophobicity. Likewise, partitioning of the unknown is driven by hydrophobicity.

=>In domain

iv.     Metabolic domain: Partitioning equilibrium can only be defined for molecules that are stable in the matrix. Although the unknown can undergo hydrolysis, the process is slow and partitioning and hydrolysis will be competing process.

=>NA.

Results from the structural analogue search b.    Structural analogues: The training data set and the validation data set have been searched for analogues using a sub-structural search for relevant fragments present in the unknown. Substructures that where searched are epoxides and biphenyls. From the initial list of hits all molecules with a significant degree of halogenation were removed. There is no molecule in the dataset where all relevant fragments are represented. All partial analogues identified were from the training set and required correction factors. Partial analogues identified in this procedure are listed in Table1. Experimental results and predicted values are listed in Table2.

 

Table1: Structural analogues identified in the training dataset.

Chemical name

Smiles

Analogues

 

Epichlorohydrin

C1(CCl)CO1

2,2-Bioxirane

C1(C2CO2)CO1

Benzophenone

c1(C(=O)c2ccccc2)ccccc1

Diphenylmethanol

c1(C(O)c2ccccc2)ccccc1

Unknown:

 

4,4'-Isopropylidenediphenol, oligomeric reaction products with 1-chloro-2,3-epoxypropane

CC(C)(c2ccc(OCC1CO1)cc2)c4ccc(OCC3CO3)cc4

 

Table2: Experimetal data and QSAR prediction for Analogues and unknown. 

Chemical name

Log Kow

Log Kocexp.

Log Kocest. (MCI)

Δ

Log Kocest. (Kow)

Δ

Analogues

 

Epichlorohydrin

0.45a

1.00a

1.00

0.00

1.08

0.08

2,2-Bioxirane

-0.28a

0.40a

0.40

0.00

0.59

0.19

Benzophenone

3.18a

2.63a

3.06

0.43

2.88

0.25

Diphenylmethanol

2.67a

2.34a

2.87

0.53

1.99

-0.35

Unknown:

 

4,4'-Isopropylidenediphenol, oligomeric reaction products with 1-chloro-2,3-epoxypropane

3.24b

 

3.81

 

2.65

 

aSource: {Schüürmann, 2006 #457}   
bREACH dossier 2010.

c.      Considerations on structural analogues: For the epoxy structure, the prediction of both methods is very close to the experimental value. For the diphenyl structure, the Log Kocis overestimated by the MCI method and slightly over- and under-predicted for the Log Kow method. Based on this finding, preference is given to the LogKow method for estimating the unknown. 3.2          The uncertainty of the prediction (OECD principle 4) Uncertainty in the value results from the notable difference in the predicted values for Log Koc using the two models with 2.65 and 3.81 for the Log Kow based and the MCI method, respectively. The difference can partially be explained by the experimental Log Kow used as an input for the model which is notably lower than the Log Kow predicted using a fragment method. Based on the data presented by the developers, approx. 95% of all predictions of Log Koc for the training fall within the range of ± 0.7 of the experimental value. This can be used to define the 95% confidence interval for theKoc of the unknown. The major contribution of the uncertainty is likely to come from the variability in the soil and sediment matrixes in the experimental data and does not reflect the uncertainty of the QSAR method.

3.3          The chemical and biological mechanisms according to the model underpinning the predicted result (OECD principle 5).  Not applicable. No mechanistic interpretation for the algorithm provided.
Validity criteria fulfilled:
yes
Conclusions:
The  Log Koc is predicted as 2.65 +/- 0.7 using the Kow method in KOCWIN v. 2.0 and Kow=3.24 as input.
Executive summary:

The KOCWIN model estimates Log Koc either from the octanol-water coefficient or from a first-order molecular connectivity index (MCI).Structural analogues as identified in the training set have been processed parallel to the unknown. The model does not define parametric or structural domain. However, the analogues identified from within the training set suggest coverage of relevant substructures present in the unknown (alkyl, benzene, epoxides).A comparison of experimental and predicted values for the analogues was performed to decide on the preferred model for the prediction. Uncertainty in the value results from the notable difference in the predicted values for Log Koc with 2.65 and 3.81 for the Log Kowbased and the MCI method, respectively. The difference can partially be explained by the lower than predicted experimental Log Kow which is used as an input value for the Log Kow method. Based on this consideration and a slightly better performance of the analogues in the Log Kow model, preference was given to value calculated with the Log Kow model.

Confidence intervals are derived based on the error histogram provided by the developer that indicate that 95% of the substances in the training dataset are predicted within +/- 0.7 units of the experimental error.

 

Name

SMILES(major component, 85%)

Log Koccalc

Koccalc

Bisphenol A diglycidyl ether

CC(C)(c2ccc(OCC1CO1)cc2)c4ccc(OCC3CO3)cc4

2.65 ± 0.7

445 (89–2228)

 

The predicted value is considered reliable for the purpose of environmental modelling and the uncertainty is comparable to the situation where experimental data for only one or a small number of different soils has been tested.

Description of key information

Log Koc = 2.65 +/- 0.7 QSAR prediction using the Kow method in KOCWIN v. 2.0 and Kow =3.24 as input. 

Key value for chemical safety assessment

Koc at 20 °C:
445

Additional information

The KOCWIN model estimates Log Koc either from the octanol-water coefficient or from a first-order molecular connectivity index (MCI). Structural analogues as identified in the training set have been processed parallel to the unknown. The model does not define parametric or structural domain. However, the analogues identified from within the training set suggest coverage of relevant substructures present in the unknown (alkyl, benzene, epoxides). A comparison of experimental and predicted values for the analogues was performed to decide on the preferred model for the prediction. Uncertainty in the value results from the notable difference in the predicted values for Log Koc with 2.65 and 3.55 for the Log Kow based and the MCI method, respectively. The difference can partially be explained by the lower than predicted experimental Log Kow which is used as an input value for the Log Kow method. Based on this consideration and a slightly better performance of the analogues in the Log Kow model, preference was given to value calculated with the Log Kow model.

Confidence intervals are derived based on the error histogram provided by the developer that indicate that 95% of the substances in the training dataset are predicted within +/- 0.7 units of the experimental error.