A multidimensional strategy to detect polypharmacological targets in the absence of structural and sequence homology.
1. Raunak Shrestha
29th November 2011
Source:
Durrant JD, Amaro RE, Xie L, Urbaniak MD, Ferguson MA, Haapalainen A, Chen Z, Di Guilmi AM, Wunder
F, Bourne PE, McCammon JA. A multidimensional strategy to detect polypharmacological targets in the
absence of structural and sequence homology. PLoS Comput Biol. 2010 Jan 22;6(1)
5. 5
Therapeutic
outcome
‘‘one gene, one drug, one disease’’
Conventional Drug Design
Therapeutic
Slide adapted from: Irene Kouskoumvekaki, From Chemoinformatics to Systems Chemical Biology
Side
Effects
Therapeutic
outcome
Polypharmacology :: multi-target drugs
Emerging Concept in Drug Design
Important Reason
for the Drug
Failure
Potential
Solution
6. • A drug that selectively binds to only one target is very rare !!!
• Many drugs interact with multiple target via a complex
network pathway
• Emergence of Drug Resistant strains of pathogens
• Resistance against a single target can be easy
• but resistance against multiple targets may be hard to achieve
• Many adverse affect of a drug is due to its interaction with
multiple target
• Some drugs may have alternative therapeutic applications
(drug repurposing)
6
7. Objective of the paper
• To identify the multiple protein receptors of a given compound
• TbREL1 is a confirmed drug target in Trypanosoma brucei
(causative agent of human African trypanosomiasis)
7
T. brucei RNA editing ligase I
(TbREL1)
NCS45208
(Compound 1)
Primary Target
Secondary
Targets ???
9. 9
NCBI blastclust
Identity threshold = 30%
overlap threshold 0.9
Queried Compound 1
in RSCB PDB
Randomly picking single chain
from each cluster as a
representative of the cluster
Select the chains having similar active
sites to the primary target TbREL1
“…. sequence order-independent profile–profile alignments
(SOIPPA) is able to detect distant evolutionary relationships
in cases where both a global sequence and structure
relationship remains obscure …. ” (Xie and Bourne, PNAS,
2008 Apr 8;105(14):5441-6)
Also included the chains having
similar active sites to that of TbREL1
Filtered only the proteins
from human or known
human-pathogen species
10. 10
In silico Docking using AutoDock 4.0
Docking cross-validated using:
• SITE data included in the published PDB file
• Examination of co-crystallized ligands bound
in native active sites
• Homology modeling : to determine the
locations of active sites for the remaining
protein chains
if Compound 1 had
a high predicted
energy of binding
Hit
If Compound 1 bind in an
identified active site of
known biochemical or
pharmacological activity
12. Secondary Target Prediction
• Compound 1 docking was performed in each of the 645
potential secondary targets of the PDBr
• both protein chains of unknown function and redundant chains
were omitted
• 87 non-redundant secondary targets were predicted
• 35 chains: known active sitses contained docked ligands
• 35 chains: alternate sites contained docked ligands
• 17 chains: could not be classified
12
Also some of the predicted secondary targets were
experimentally verified in wet-lab
13. Predicted Human Proteins
(secondary targets)
• 12 were Human Protein (out of 35 predicted secondary targets )
13
Binding
Energy
Active site
similarity
Structural
similarity
Neither FATCAT nor CLUSTALW2 could predict any similarity between most of
the primary target and secondary target but SOIPPA algorithm along with
Docking confirmed as a potential secondary targets
14. Predicted Bacterial and Parasitic
Pathogens Proteins (secondary targets)
• 23 were bacterial and parasitic pathogens protein (out of 35 predicted
secondary targets )
14
Binding
Energy
Active site
similarity
Structural
similarity
Neither FATCAT nor CLUSTALW2 could predict any similarity between most of
the primary target and secondary target but SOIPPA algorithm along with
Docking confirmed as a potential secondary targets
15. Conclusion
• A good computational pipeline to predict the off-targets
(secondary targets) of a compound.
• Can give significant insight over the system-biology of
druggable genome
• Also give valuable insight over the possible side-effects of a
drug
• Even in the absence of sequence homology, the pipeline can
predict the off-targets.
15
16. Critique
• FATCAT : a popular structural alignment tool
for proteins
• SOIPPA algorithm + Docking predicted
secondary targets when even FATCAT and
CLUSTALW2 could not !!!
• The pipeline seems to be very efficient to
detect secondary targets
16
1
• Performed Experimental (wet-lab) validation
of the secondary targets
• Most of the bioinformatics papers does not seem
to do so !!!
2
17. Critique
• Randomly selected the chains from the cluster
• Can a single randomly selected protein from a cluster
be a representative of the cluster ?
• Can be accepted if there is conversation within the active site
residues (but this information is not mentioned in the paper)
• Taking a consensus sequence could be an alternative
• If so generating the structural information would be very
difficult for a consensus sequence
17
1
19. Limitations
19Limited structural coverage of the given proteome
This will seriously limit the algorithm ability to predict secondary-targets
Number of structures in the PDB from 1972 - 2010. Image courtesy of the RCSB Protein Data Bank.
SITE records specify residues comprising catalytic, co-factor, anti-codon, regulatory or other essential sites or environments surrounding ligands present in the structure
Flexible structure AlignmenT by Chaining AFPs (Aligned Fragment Pairs) with Twists (FATCAT)
Flexible structure AlignmenT by Chaining AFPs (Aligned Fragment Pairs) with Twists (FATCAT)
Flexible structure AlignmenT by Chaining AFPs (Aligned Fragment Pairs) with Twists (FATCAT)