Model Applicability Domain in OASIS QSAR Models

1
Model Applicability Domain in OASIS QSAR Models
Laboratory of Mathematical Chemistry University
Prof. As. Zlatarov, 8010 Bourgas, Bulgaria
2
Applicability domain of (Q)SAR models
Dimitrov S., et al. (2005) A stepwise approach
for defining applicability domain of SAR and QSAR
models, J Chem Inf Model, 45, 839-849.
S
Parent
High level
W
Low level
S
Confidence of the obtained predictions
Skin sensitization model
Applicability domain
3
Applicability domain of (Q)SAR models
I. General requirements
II. Structural domain
III. Mechanistic domain
IV. Metabolic simulator domain
4
Applicability domain of (Q)SAR models
I. General requirements
Screened chemicals S
Domain of General Requirements, DGR
SGR
Molecular weight, MW Hydrophobicity, log
Kow Water solubility, SW

ST
SMS
5
Applicability domain of (Q)SAR models
II. Structural domain
Screened chemicals S
Partitioning of molecule into atom-centered
fragments
SGR
ST training set
SC correctly predicted chemicals SC ? ST

ST
DT
ST
SC
DSD
SMS
6
Applicability domain of (Q)SAR models
Defining the structural domain accounting for the
second neighbours of N-atom
Extracted substructures
Chemical
Chemical
does not belong to the structural domain of N-atom
belongs to the structural domain of N-atom
7
Applicability domain of (Q)SAR models
Defining the structural domain accounting for the
second neighbours of N-atom
Extracted substructures
Chemical
Chemical
does not belong to the structural domain of N-atom
belongs to the structural domain of N-atom
8
Applicability domain of (Q)SAR models
III. Mechanistic domain
Screened chemicals S
The most multifarious and insusceptible to
standardization
SGR

• model structure
• mathematical formalism
• computational method
• accepted hypotheses
• etc.

ST

• Domain of functional (reactive) groups, DFG
• Domain of explanatory variables DEV

SMS
9
Applicability domain of (Q)SAR models
III. Mechanistic domain

• Domain of functional (reactive) groups, DFG

Reliability of functional groups (probability for
correct predictions)
and are the numbers of
successful and unsuccessful applications of ith
functional group
TFG is the threshold for the significance of
,
10
Applicability domain of (Q)SAR models
III. Mechanistic domain
2. Domain of explanatory variables, DEV
Parameter interpolation space Local performance
of the model
11
Applicability domain of (Q)SAR models
III. Mechanistic domain
2. Domain of explanatory variables, DEV
Parameter interpolation space
Excluding sparsely populated peripheries of the
parametric interpolation domain
80 of descriptor interpolation space DEV
12
Applicability domain of (Q)SAR models
III. Mechanistic domain
2. Domain of explanatory variables, DEV
Local performance of the model
Probability the model prediction to be correct
in the neighborhood of the screened
chemical Probability the descriptors of the
training chemicals to be in the neighborhood of
the screened chemical
13
Applicability domain of (Q)SAR models
IV. Metabolic simulator domain
Screened chemicals S

• Reliability of the ith transformation
• Reliability of a predicted metabolite
• Reliability of the generated metabolic map for
  kth parent

SGR

ST
SMS
14
96 external chemicals
Predictability,
Application of stages I - IV of model domain
100
100
83
80
79
69
49
49
33
29
25
26