Transcript In-silico design of HIV-1 protease inhibitors

Designing a Tri-Peptide based HIV-1
protease inhibitor
Presented by,
Sushil Kumar Singh
IBAB,Bangalore
Submitted to
Dr. Indira Ghosh
AstraZeneca India Research Center, Bangalore
Objective
The objective of my project was to come up with a certain
number of lead molecules which can be potential HIV1
Protease inhibitors.
Based on the assignments I have tried to come up with
different ways of designing drug molecules for the same
target.
In the end I expected to have a library of molecules having a
high possibility of activity against the target.
This information is vital for the future of rational drug
designing, leading to more effective drugs with minimal side
effects.
Flowchart of the path taken for the project
Protocol
●
Built the model using Homology from InsightII.
Analog-Based Drug Designing
●
●
●
Used Analog Builder to generate both product and reagent
based libraries using a scaffold from published literature.
The library was annotated using the ' Lipinski's Rule of Five '.
The resulting compounds were tested for docking score using
Ligand-Fit and then their activity was predicted using Hypogen
as well as QSAR.
Structure-Based Drug Design
●
Structure of active site of receptor were analysed and analog
were generated ,but unfortunately we got some negative score.
Pharmacophore-Based Drug Design
This was achieved using 3 different methods:
1.A pharmacophore was built based on the structure of receptor.
2.Pharmacophore based on the common feature of known drugs.
(Hip-Hop)
3.Based on the activity of the known potential hiv protease
inhibitors (training set of data)- Hypo-gen
- Using the best hypothesis from Hip-Hop the generated
compounds were analysed.
- The best hypothesis resulting from Hypo-gen was used to predict
the activity of the lead molecule.
QSAR Activity prediction - The activity of the lead compounds
generated earlier was predicted using QSAR+.
Analysis
●
●
The analog library which was generated, many of the compounds
were not following the ' Lipinski's Rule of 5'. On analysis I found
that the fragments which I selected for library building were of
high mol wt, more no. of HB donor/acceptor and in some cases
no. of rotatable bonds were also more. To avoid this I generated
library using fragments with low mol wt. And less no. of
rotatable bonds.
Hypothesis generation:
For a hypothesis to be good
- the range between Fixed cost and Null cost should be high.
( F.c- N.c > 85 )
- the total cost should be closer to the fixed cost.
- The Config. Cost factor should be less than 17.
Hypothesis Analysis
I faced difficulty in deciding which hypothesis to take out of 10.
Because many of the hypothesis were showing same type of
variations with their values.
To come out with the best hypothesis I did clustering, that also
didn't give good result as most of the hypothesis were of same
nature.
Then we analysed the training set of data, which we used for
hypothesis generation and found that most of the fragment were
of common character, which might be the reason for not getting
much variation.
To proceed further, we took one hypothesis, but we couldn't
validate our result using this hypothesis as it is not following the
criteria of config and costs.
Score comparision
Null Hypothesis:
dumping score for the null hypothesis:
Total cost=57.9634 RMS=1.7127 correl=0
Cost components: Error=57.9634 Weight=0 Config=0 Mapping=0
Tolerance= 0
Hypothesis Taken:
Total cost=64.0193 RMS=0.634962 correl=0.937511
Cost components: Error=42.7823 Weight=1.58709 Config=19.6499
Tolerance=0
Fixed Cost:Total cost=61.1381 RMS=0 correl=0
Cost components: Error=40.3633 Weight=1.12491 Config=19.6499
Tolerance=0
Activity (Ic 50)
QSAR
Molecules for the training set of data were generated into
catalyst and imported to Cerius2. They were added to the
study-table.
Descriptors taken:
Topological descriptors
Hosoya index, Zagreb index, Chi index, Winner index
etc.
Fragments constant descriptors
HB acceptors, HB donor etc.
Charge desriptors
Charge, dipole Apo l etc.
Every descriptor adds a dimension to the chemical space,
to reduce the dimensionality without loosing any information
we did PCA ( Principal Component Analysis).
Using GFA we predicted the activities of the compound which we
QSAR analysis
●
By QSAR analysis I tried to find contribution of active fragments in
the activity of the compound.
While analysing the QSAR equations I found many terms with
negative sign which were not good for the activity, so I substituted
the groups to nullify the effect.
By making the substitution I noticed that activity of the compound is
changing.
e.g
If we replace ester group by an amide group we found activity
increased by 300 fold.
On introducing HB donor or acceptor, the activity of the compound
decreased as we found that replacement of methyl group with
hydroxyl group led to 2500 times lower activity.
Conclusion