Transcript Aliphatic QSAR

QSARs to Predict Extent of
Drug Biotransformation in
Humans
Na’ngono Manga, Judith C. Duffy,
Phil H. Rowe, Mark T.D. Cronin,
School of Pharmacy and Chemistry
Liverpool John Moores University
Introduction
• Project failures mainly attributed to
pharmacokinetic problems
• Increasing effort placed into
forecasting pharmacokinetic
properties of drugs
• Most work has focussed on
absorption, transporters, metabolism
enzymes and blood brain barrier
Aims of Study
• To model extent of drug
biotransformation in humans as a
composite process, using urinary
excretion of unchanged drug
• To develop a transparent QSAR
model in rational manner accepting
non-linear nature of data
Methods
• Source of Data
• Training set: 160; Test set: 40 drugs
• Molecular weight capped at 500
• Response data
• Cumulative amount of unchanged drug
excreted in the urine expressed as percent of
i.v. dose
• Data categorised into low, medium, high
• No boundary values assigned a priori
Methods
• Physico-chemical properties calculated:
• log P, pKa, log D
• Structural and topological parameters
• Molecular orbital (AM1) properties
• Statistics:
• Stepwise LDA
• Recursive partitioning
Urinary Excretion against log D6.5
< 0.3
> 0.3
urinexcr
100
50
0
-10
0
logd65
10
Local Models of the Data
• Drugs with log D6.5 > 0.3 removed
• Attempt to build local models on
remaining classified data:
• 1- Low vs. high urinary excretion
compounds
• 2 - Low vs. medium
• 3 - Medium vs. high
Low (< 10%) vs. High (> 75 %)
Urinary Excretion
W = 1.11 {7.07 + 27.7 Sacid + 21.7 SHBD
– 23.6 SOH} – 17 SQ7 + 15.5
urinexcr
100
50
0
0
50
W
100
Modelling Whole Data Set
Compounds with log D6.5 > 0.3
assumed to have low urinary excretion
< 0.3
> 0.3
urinexcr
100
50
0
-10
0
logd65
10
Discriminant Function Developed
for Drugs with log D6.5 < 0.3
urinexcr
100
50
0
0
50
100
W
Classifications from LDA were
Subjected to Recursive Partitioning
For compounds with W  47 satisfactory decision
tree resulted:
Urinary Excretion  38% if Total Energy > 15,
or if HB < 2, or If –27919 < EE < -19320, Or If
IP < 8.96 or IP > 9.78
Otherwise Urinary Excretion > 38%
Decision Tree for Whole Data Set
Log D6.5
Discriminant
Function
> 0.3
> 47
Excretion < 25%
Excretion > 38%
Recursive Partitioning
Excretion  38%
Excretion > 38%
Classification and Validation Using
Test Set
Decision Level
Ratio of correct
Percentage
Compounds
misclassified as
extensively
metabolised
Decision
Ratio Percentage Misclassified
Level of correct of correct
as low
Level 1
Level 2
Level3
Overall
Level 1
Level 2
Level3
Overall
Level 1
Misclassified as
medium/high
70/73 compounds
0
31/36
5
70/73 drugs
Level 2 31/36
45/51
Level3146/160
Overall
TRAINING SET
95%
3
95%
3 drugs
86% 86%
00
88%
5
59
91% 88%
TEST SET
91%
9
84%
3 drugs
93%
0
66%
1
85%
4
Compounds
misclassified as
moderately/poorly
metabolised
16/19 drugs
14/15
4/6
34/40
45/51
146/160
1
6
0 drugs
5
1
6
0 drugs
1
1
2
Discussion
• Hybrid metabolism data can be
modelled adequately
• Model uses descriptors related to
metabolism
• Weakness in drugs with medium
urinary excretion and drugs with long
half-lives