ABTox: Acute Toxicity

1
AB/Tox Acute Toxicity
C-SAR/QSAR Approach in Predicting Acute Toxicity
D. Zmuidinavicius, L. Steponeniene, R.
Didziapetris, P. Japertas, L. Riauba, A.
Petrauskas Pharma Algorithms, Inc.,
www.ap-algorithms.com
Click to view
Strategies of Analysis
LD50 Data Sets
Expert Knowledge
Classification SARs
Toxicophores/QSARs
ADME Effects
Algorithms
Validation
Applications
QSAR02 Poster
2
Strategies of LD50 Analysis
Depend on the complexity/diversity of

• chemical structures

• biological mechanisms

• One-step QSAR analysis

• Result statistical algorithms

• Examples Topkat, ToxSys, ToxAlert

Tox-Data

• Multi-step analysis

• Explains ADME/Tox effects

• Result knowledge algorithms

• Example AB/Tox

3
LD50 Data Sets
RODENT TOXICITY
FISH (F. MINNOW)
4
Expert Knowledge

• Many types of toxicity predictions are based on
  expert knowledge

• Examples genotoxicity, carcinogenicity,
  irritation, etc.

• Not much expert knowledge is available for LD50
  predictions

• The reason - complexity of ADME/Tox effects in
  live organisms

• LD50 may increase

• Many different in vivo systems

• or decrease

• Many thousands of compounds

• Classification SAR techniques

5
Classification SAR
1. Classification SAR (C-SAR) based on Recursive
Partitioning (RP)

• Identifies the most important factors that
  affect LD50 values

• Groups compounds into classes with possibly
  different mechanisms

2. Descriptors include

• Functional groups

• Interactions (pairs of groups)

• MW, MV, tPSA

• Branched chains of atoms

• Phys-chem properties

• LogP, H-Bonding

3. Example of C-SAR analysis for 30,000 compounds
with LD50 values in mouse, intraperitoneal
administration

• Aqueous Solubility

• pKa, Ion Fractions

4. The identified factors serve as a basis for

• Defining new toxicophores

• Building rule-based algorithm(s)

• Conducting class-specific QSAR analyses

6
Toxicophores/QSARs
C-SAR analysis identified many substructures that
cause specific toxic effects
Most of these effects agree well with independent
knowledge from literature
The identified effects serve as the basis for the
class-specific QSAR analysis
-Log(LD50/MW) ? Toxicophores ? Other
Fragments ? Interactions

• Similar tox-effects can be described by a single
  model

• Different tox-effecits require different QSAR
  models

7
ADME Effects
Comparison of the results in different
animal/administration systems

• Mouse and Rat have very similar ADME effects

• LD50 in mouse and rat are interchangeable
• (under the same type of administration).

• Oral administration differs from
• other types of administration

• Example

gtNlt cannot cross intestinal membrane
so it is non-toxic in oral administration,
but highly toxic in non-oral administration.
(Certain exceptions exist.)
8
Algorithms
AB/Tox algorithm works as follows

• Similarity to toxicophores is checked

• Tox-specific QSAR models are used

• Automated decision regarding
• compounds toxicity is produced

Compare to the Statistical algorithms

• QSAR models disregard
• ADME/Tox mechanisms

• Similarity analysis is not
• tox-specific

• Made only after LD50
• has been calculated!

9
AB/Tox Validation

• Correlation of exp. vs. calc. pLD50 values
  (mM/Kg)

Mouse, intraperitoneal N30,242, R0.83, SD0.37
Validation N3,338, R0.71, SD0.47
Mouse, intravenous N17,272, R0.91, SD0.30
Validation N1,919, R0.77, SD0.47

• Qualitative identification of toxic compounds

False negatives 20 - 25
(toxic compounds, predicted as non-toxic)
False positives 7 - 8
(non-toxic compounds, predicted as toxic)
The highest reliability of HTS
screening available today
10
Applications
1. Estimation of LD50 in six animal systems
2. Virtual screening of large databases
3. Optimization of new lead compounds
(combined with ADME filters)
11
QSAR02 Poster
1st Slide ?