QSAR Application Toolbox: Step 12: Building a QSAR model

1
QSAR Application ToolboxStep 12 Building a
QSAR model
2
Objectives

• This presentation demonstrates building a QSAR
  model for predicting acute toxicity to
  Tetrahymena pyriformis of aldehydes. The
  presentation addresses specifically
• predicting acute toxicity for a target chemical
• building QSAR model based on the prediction
• applying the model to other aldehydes
• exporting the predictions to a file.

3
The Exercise

• This exercise includes the following steps
• select a target chemical Furfural, CAS 98011
• extract available experimental results
• search for analogues
• estimate the 48h-IGC50 for Tetrahymena pyriformis
  by using trend analysis
• improve the data set by either
• subcategorising by Protein binding mechanisms,
  or
• assessing the difference between outliers and the
  target chemical
• evaluate and save the model
• Use the model to display its training set,
  visualize its applicability domain and perform
  predictions.

4
Chemical Input

• After launching the Toolbox, select the Flexible
  Track.
• This takes you to the first module, which is
  Chemical input.
• Enter the target chemical by its CAS number
  (98-01-1)

5
Select target chemical Furfural, CAS 98011
6
Substance Information
7
Profiling the Target Chemical

• Select the Profiling methods you wish to use by
  clicking on the box before the name of the
  profiler.
• For this example check all mechanistic methods.
• Click on Apply.

8
Profiling
9
Target interaction with proteins
Double clicking shows profiling scheme
The chemical could interact with protein by
Schiff-base formation.
10
Target interaction with proteins
11
Endpoints

• Endpoints refer to the electronic process of
  retrieving the environmental fate, ecotoxicity
  and toxicity data that are stored in the Toolbox
  database.
• Data gathering can be executed in a global
  fashion (i.e., collecting all data of all
  endpoints) or on a more narrowly defined basis
  (e.g., collecting data for a single or limited
  number of endpoints).

12
Extracting endpoint values
13
Redundancy table Reports for same endpoint
values across databases
14
Reproducing endpoint value
In this exercise we will build a QSAR model to
estimate the following endpoint
Ecotoxicological Information Aquatic
Toxicity Protozoa Tetrahymena
pyriformis IGC50 48h
15
Defining a Category
The initial search for analogues is based on
structural similarity, in this example - US
EPA categorization
16
Category Definition
17
Set Category Name
18
Analogues

• The data is automatically collated.
• Based on the defined category (Aldehydes US EPA
  categorisation) 274 analogues have been
  identified.
• These 274 compounds along with the target
  chemical form a category (Aldehydes), which can
  be used for data gap filling (see next slide).

19
Analogues
20
Extracting experimental results for analogues

• Highlight the 274 Aldehydes (US EPA
  categorisation).
• The inserted window entitled Read Data?
  appears (see next slide).
• Click OK.

21
Extracting experimental results for analogues
22
Extracting experimental results for analogues
23
Applying Trend-analysis

• Move to the module Filling data gap
• Open the data tree to
• Ecotoxicological information
• Protozoa
• Tetrahymena pyriformis
• IGC50
• 48 h
• Highlight the data endpoint box under the target
  chemical.
• It contains already an experimental result, which
  we are going to reproduce by trend analysis.
• Next with the trend analysis box highlighted,
  click Apply (see next slide).

24
Apply Trend-analysis
25
Results of Trend-analysis
26
Interpreting the Trend-analysis

• The resulting plot outlines the available
  experimental results of all analogues (Y axis)
  according to a default descriptor Log Kow (X
  axis).
• The RED dot represents the target chemical.
• The BLUE dots represent the experimental results
  available for the analogues.
• The GREEN dots represent the analogues belonging
  to a different subcategory (see following slides).

27
An Accurate Trend Analysis of the Data set (1)

• In this example, the mechanistic properties of
  the analogues are not consistent.
• Subcategorization can be performed based on
  protein binding mechanisms. This is the second
  stage of analogue search - requiring the same
  interaction mechanism.
• Acute effects are indeed associated with
  interaction of chemicals with lipid cell
  membrane, i.e. with protein binding.
• Chemicals with a different protein binding
  mechanism compared to the target chemical will be
  removed.

28
Subcategorization

• To improve the data by subcategorizing, follow
  these steps
• Click on Subcategor.
• Select Protein binding from the Grouping methods
  list.
• All chemicals which have a potential protein
  binding mechanism different from the target
  chemical are highlighted (GREEN dots)
• Click on Remove.

29
Subcategorization
30
Result after Subcategorization
31
An Accurate Trend Analysis of the Data set (2)

• The chemicals which differ from the target are
• Michael type nucleophilic addition (23)
• No binding (48)
• Nucleophilic addition to azomethynes (1)
• Nucleophilic substitution of haloaromatics (1)
• Another way for refining the data set is to ask
  what makes the obvious outliers different from
  the target.

32
Subcategorization

• Right-Click on any of the outlying results from
  the analogues (BLUE dots)
• Select Differences to target from the menu
• Select Protein binding from the Grouping methods
  list
• Click on Remove (see next slide)

33
Subcategorization
34
Result after Subcategorization
35
QSAR Model evaluation

• To assess the model accuracy use
• - Adequacy (predictions after leave-one-out)
• - Statistics
• - Cumulative frequency

36
QSAR Model evaluation
37
QSAR Model evaluation
38
QSAR Model evaluation
The residuals abs (obs-predicted) for 95 of
analogues are comparable with the variation of
experimental data.
39
Saving the Derived QSAR Model

• To save the new regression model follow these
  steps
• - Click on Save model button
• - Enter the model name Acute tox
• - Click on OK and
• - Accept the value

40
QSAR Model evaluation
41
Apply QSAR model

• The derived model can be used to
• List training set chemicals
• Right-click on the QSAR model Acute tox
• Select training set from the context menu
• Visualize whether a chemical is in the
  applicability domain of the model
• In the data matrix highlight the empty cell of
  one of the analogues (e.g. chemical no 2 in the
  matrix) for the endpoint 48-h IGC Tetrahymena
  pyriformis
• Right-click on the QSAR model Acute tox
• Select Display domain
• Perform predictions for the chemicals in the
  matrix.
• Right-click on the QSAR model Acute tox
• Select Predict endpoint and All Chemicals in
  domain

42
Apply QSAR model Training set
43
Apply QSAR model Visualize whether a chemical is
in the applicability domain of the model

• The chemical is an aldehyde as required by the
  model. It can react with protein by Schiff-base
  formation and does not react to protein by any of
  the eliminated mechanisms
• Michael-type nucleophilic addition
• No binding
• Nucleophilic addition to azomethynes
• Nucleophilic substitution of haloaromatics
• Another requirement is Log Kow to be gt0.3210
  and lt 4.75. The last requirement is slightly
  violated (Log Kow 4.87) and therefore the
  chemical is outside of the applicability domain
  of the model.

44
Apply QSAR model Visualize whether a chemical is
in the applicability domain of the model
45
Apply QSAR model Perform predictions
46
Apply QSAR model Perform predictions
47
Export QSAR results

• The predictions for the chemicals in the matrix
  can be exported into a text file.
• In the data tree right-click on 48 h (for the
  endpoint IGC50 for Tetrahymena pyriformis) and
  select Export endpoint data from the menu.

48
Export QSAR results
click right button
49
Export QSAR results
50
Export QSAR results
51
Export QSAR results

• The resulting text file can be loaded into a
  spreadsheet and further analysed.